RESEARCH ARTICLE Open Access

Genome-wide association mapping ofpartial resistance to Aphanomyces euteichesin peaAurore Desgroux1,2, Virginie L’Anthoëne1,10, Martine Roux-Duparque3,11, Jean-Philippe Rivière1,4, Grégoire Aubert2,Nadim Tayeh2, Anne Moussart4,5, Pierre Mangin6, Pierrick Vetel1,4, Christophe Piriou1,4, Rebecca J. McGee7,Clarice J. Coyne8, Judith Burstin2, Alain Baranger1,4, Maria Manzanares-Dauleux1,9, Virginie Bourion2

and Marie-Laure Pilet-Nayel1,4*

Abstract

Background: Genome-wide association (GWA) mapping has recently emerged as a valuable approach for refiningthe genetic basis of polygenic resistance to plant diseases, which are increasingly used in integrated strategies fordurable crop protection. Aphanomyces euteiches is a soil-borne pathogen of pea and other legumes worldwide,which causes yield-damaging root rot. Linkage mapping studies reported quantitative trait loci (QTL) controllingresistance to A. euteiches in pea. However the confidence intervals (CIs) of these QTL remained large and wereoften linked to undesirable alleles, which limited their application in breeding. The aim of this study was to use aGWA approach to validate and refine CIs of the previously reported Aphanomyces resistance QTL, as well asidentify new resistance loci.

Methods: A pea-Aphanomyces collection of 175 pea lines, enriched in germplasm derived from previously studiedresistant sources, was evaluated for resistance to A. euteiches in field infested nurseries in nine environments andwith two strains in climatic chambers. The collection was genotyped using 13,204 SNPs from the recentlydeveloped GenoPea Infinium® BeadChip.

Results: GWA analysis detected a total of 52 QTL of small size-intervals associated with resistance to A. euteiches,using the recently developed Multi-Locus Mixed Model. The analysis validated six of the seven previously reportedmain Aphanomyces resistance QTL and detected novel resistance loci. It also provided marker haplotypes at 14consistent QTL regions associated with increased resistance and highlighted accumulation of favourable haplotypesin the most resistant lines. Previous linkages between resistance alleles and undesired late-flowering alleles for drypea breeding were mostly confirmed, but the linkage between loci controlling resistance and coloured flowers wasbroken due to the high resolution of the analysis. A high proportion of the putative candidate genes underlyingresistance loci encoded stress-related proteins and others suggested that the QTL are involved in diverse functions.

Conclusion: This study provides valuable markers, marker haplotypes and germplasm lines to increase levels ofpartial resistance to A. euteiches in pea breeding.

Keywords: Root rot, Plant disease resistance, GWAS, Pea (Pisum sativum), Quantitative trait loci, Marker haplotype,Candidate genes

* Correspondence: Marie-Laure.Pilet@rennes.inra.fr1INRA, UMR IGEPP 1349, Institut de Génétique et Protection des Plantes,Domaine de la Motte au Vicomte, BP 35327, 35653 Le Rheu Cedex, France4PISOM, UMT INRA/Terres Inovia, UMR IGEPP 1349, Domaine de la Motte auVicomte, BP 35327, 35653 Le Rheu Cedex, FranceFull list of author information is available at the end of the article

© 2016 Desgroux et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Desgroux et al. BMC Genomics (2016) 17:124 DOI 10.1186/s12864-016-2429-4

BackgroundResearch interest in polygenic resistance to plant dis-eases has increased worldwide in the past ten years tomeet the challenge of sustainable agriculture. Recom-mendations to reduce chemical inputs and the frequentbreakdown of major resistance genes in plants have en-couraged the integration of polygenic resistance into cul-tivars of many crops. However, breeding strategies forpolygenic resistance, which is controlled by many genes,have not been as widely developed as for monogenic oroligogenic resistance (controlled by one or few genes, re-spectively) [1]. Breeding schemes for polygenic traits arecostly and time-consuming. The effects of QuantitativeTrait Loci (QTL) controlling resistance are not alwaysconserved in different genetic backgrounds and environ-ments and markers tightly linked to resistance loci havealso often been lacking. Further research is needed tovalidate QTL effects, reduce their confidence intervals(CIs) and identify their underlying causal genes [2, 3], toencourage and optimize QTL use in breeding [4].The identification of plant resistance QTL has broadly

been explored using linkage mapping populations de-rived from crosses between two parental genotypes [1].With the decrease in genotyping costs and the massivedevelopment of markers in the recent past years,genome-wide association (GWA) studies, are becomingcommon approaches to detect natural variation under-lying complex traits, especially polygenic resistance tomajor diseases, in a large range of crop species [5–7], in-cluding legumes [8, 9]. The advantages of GWA studiescompared to bi-parental linkage mapping include accessto wider genetic diversity, higher recombination ratesdue to the evolutionary history of the species, and thussubstantially refined genomic regions associated withtrait variations [5, 10]. Accurate marker density forGWA studies depends on the rate of linkage disequilib-rium (LD) decay and should be higher in species with arapid LD decay (a few kb, such as in maize) than inthose with a slow LD decay (~100 kb, such as in rice)[5]. Segura et al. [11] recently proposed a Multi-LocusMixed Model (MLMM) approach, in order to improveGWA studies precision and power of detection, and itwas successfully applied [12]. The reduction of CIs ofgenomic regions associated with traits of interest,opened the possibility of identifying haplotypes formarker assisted selection (MAS) [8, 13] and pinpointinginteresting candidate genes underlying QTL [14, 15].However, the GWA approach has also been reported tohave poor power to detect rare alleles associated withthe trait of interest, leading to missing heritability [5,16], and complementarity between linkage and GWAapproaches has been underlined [7, 17]. Multi-parentaldesigns, including Nested Association Mapping (NAM)[18, 19], Multi-parent Advanced Generation Inter-Cross

(MAGIC) [20, 21] and breeding line populations [22, 23]were proven to efficiently increase power of GWA stud-ies to detect rare variants, for which rates are increasedby selection of rare-allele-carrier parental lines [7, 16].Dry pea (Pisum sativum) is the third most important

pulse crop worldwide [24], for which yield has been un-stable for the past decades, mainly due to biotic and abi-otic stresses. One of the most damaging biotic stressesof peas is Aphanomyces root rot due to Aphanomyceseuteiches [25]. The soil-borne root pathogen, first de-scribed in 1925 [26], has been mainly reported as a yieldlimiting factor in the United States of America (USA)and Europe for more than twenty years [27–29], andmore recently in Canada [30]. Two main pathotypes ofA. euteiches were described by Wicker and Rouxel [28],including pathotype I predominant in France and patho-type III detected in some regions of the USA (Onfroy etal., personal communication). Both pathotypes causehoney brown necrosis symptoms on pea roots and epi-cotyls, resulting in dwarfism, foliage yellowing and thendeath of plants in highly infested fields. Increasing yieldloss due to A. euteiches in dry and green pea productionhas been noted in Western Europe due to short crop ro-tations of susceptible pea varieties and the long lifespanof oospores [27]. The development of resistant cultivarshas been considered as a major objective for the pasttwo decades in France, as only prophylactic and crop-ping methods are available to manage the disease. Pealines partially resistant to A. euteiches were identifiedfrom germplasm screening and breeding programs con-ducted in the USA [31–35], and more recently, from aFrench germplasm screening program of approximately1900 Pisum lines [36]. The most resistant lines were in-tegrated into crossing programs to develop breedinglines [37, 38], recombinant inbred lines (RILs) [39–43]and near-isogenic lines (NILs) [44]. Breeding lines withincreased levels of resistance to A. euteiches were se-lected in a phenotypic recurrent selection-based breed-ing program developed by GSP (Groupement desSélectionneurs de Pois Protéagineux, France) [37, 38].RILs have also been used for discovery of Aphanomycesresistance QTL [39–41, 43]. A total of 27 meta-QTLwere identified on a consensus genetic map from fourRIL populations [43]. Eleven of them, corresponding toseven genomic regions, were detected on six of theseven pea linkage groups (LGs), with high consistencyover locations, years, isolates and populations [43].Marker assisted back-crossing was used to introgresseach of the seven consistent genomic regions into one ofthe susceptible RIL parents and two main spring andwinter pea varieties. The resulting NILs were used tovalidate individual or combined major resistance QTLeffects [44]. Lavaud et al. [44] considered large QTL in-tervals for NIL creation, which brought undesirable

Desgroux et al. BMC Genomics (2016) 17:124 Page 2 of 21

morphological (coloured flowers, normal leaves) or de-velopmental (late flowering) alleles linked to resistancealleles at several QTL.Massive numbers of Single-Nucleotide Polymorphism

(SNP) markers were recently developed from whole gen-ome cDNA (coding deoxyribonucleic acid) [45–47] orgenomic sequencing of pea lines [48, 49]. A GenoPea Infi-nium® BeadChip was developed by Tayeh et al. [49], con-taining 13,204 SNPs, all located in gene-contextsequences. Pea diversity panels, especially the USDA(United States Department of Agriculture) core collectionand the INRA (Institut National de la Recherche Agrono-mique) reference collection, were used to determine asso-ciations between low to medium density genetic markers(137–1233) and traits of interest [9, 50–52]. However, onlya few sources of resistance to A. euteiches were identifiedin these collections (Pilet-Nayel et al., unpublished data),as was also found in larger Pisum screening programs forAphanomyces resistance [34, 36].The aim of this study was to validate and refine the

CIs of previously reported Aphanomyces resistanceQTL, as well as identifying new resistance loci, using aGWA approach. A novel panel, enriched in pea linespartially resistant to A. euteiches from gene pools previ-ously studied, was designed including breeding andgermplasm lines [37, 38, 43]. The recent GenoPea Infi-nium® BeadChip was used for high-density SNP genotyp-ing of the collection [49]. The GWA study detectedSNPs and LD blocks associated with Aphanomyces re-sistance from data collected in nine field environmentsand two strains under controlled conditions. The gen-omic positions and CIs of resistance loci detected byGWA study were compared to those previously identi-fied by linkage analysis [41, 43]. The GWA study alsoidentified loci associated with morphological and devel-opmental traits, in order to analyse their linkages withAphanomyces resistance loci. The GWA study allowedmarker haplotypes and putative candidate genes to beidentified, for future pyramiding of resistance alleles inbreeding and to investigate the molecular basis of poly-genic resistance.

MethodsPlant materialA collection of 175 Pisum sativum lines, referred to asthe “pea-Aphanomyces collection", was used for associ-ation mapping. It includes cultivars, breeding lines andgermplasm lines, supplied by public and private pro-grams in legume genetics research and breeding(Additional file 1):

(i) One hundred and one breeding lines, partiallyresistant or susceptible to A. euteiches, originatingfrom the Aphanomyces recurrent selection-based

breeding program conducted by GSP [37, 38]. Thebreeding scheme included four crossing programs(namely AeA95, AeB97, AeC98 and AeD99),established from 1995 to 1999. Parental lines of thecrosses included partially resistant germplasm linesconfirmed in French field conditions (PI180693, 90–2131 and 552) [31–34] and dry spring or winter peacultivars with good agronomic values but susceptibleto the pathogen [37, 38]. Simple or double crossesbetween multiple parents were designed in theAeA95, AeB97 and AeC98 programs.. The best F1hybrids from the three programs were selected basedon resistance levels of their inbred progenies (F5-F6).The selected F1 hybrids were then inter-crossed inthe AeD99 program which was divided into twoparts. The first part (AeD99QU) was generated fromtwo crosses and aimed at developing pea lines withimproved resistance levels for breeding purposes.The second part (AeD99OSW) was generated fromthree other crosses and was conducted as a diver-gent selection program of resistant and susceptiblelines for QTL validation studies. In each of the fourcrossing programs, selection for resistance wasconducted from F3 or F4 generations, alternatively ingrowth chamber and in French field nurseries.Resistance levels and agronomic values of the bestF7 and F8 lines were then assessed in infested fieldtrials and in healthy field conditions.

(ii)Twenty RILs from INRA and USDA researchprograms selected for Aphanomyces resistance andfor morphological and developmental traits. TheseRILs included: (iia) four lines from the Baccara xPI180693 RIL population [41], (iib) eight lines fromthe Baccara x 552 RIL population [41], and (iic)eight lines from the Dark Skin Perfection (DSP) x90-2131 RIL population [42, 43].

(iii) Eighteen parental lines of RIL populations used forQTL analysis of Aphanomyces resistance [39–41, 43],and/or of the Aphanomyces recurrent selection-basedbreeding program conducted by the GSP. These linesincluded the six pea differential genotypes defined byWicker et al. [29] to verify disease severity and strainpathotypes in pathogenicity tests.

(iv) Twenty lines representing new sources of resistance,selected from the INRA screening program ofAphanomyces resistance sources previouslyconducted [36]. The new sources of resistanceselected included pea breeding and germplasm linesfrom Aphanomyces resistance screening programs inthe USA and from collections of genetic resourcecenters in France, the Netherlands, Russia and theUSA (Pilet-Nayel et al., unpublished).

(v)Sixteen spring or winter pea varieties currently orformerly grown in Europe.

Desgroux et al. BMC Genomics (2016) 17:124 Page 3 of 21

The pea-Aphanomyces collection thus combined dif-ferent genetic sources of resistance, especially originatingfrom three major resistant pea lines effective in France(90–2131, 552 and PI180693) and analysed in previousQTL studies [43]. The collection included plant materialdescribed both for different end-uses (food, feed or fod-der peas) and for different sowing times (spring and win-ter peas). Morphological descriptions (seed type, foliagetype and flower colour) and available pedigree informa-tion of the lines are presented in Additional file 1.

PhenotypingThe collection was evaluated for Aphanomyces resist-ance in inoculated controlled conditions (CC) andinfested fields. In CC, resistance tests were performed in2009 on all the lines but the 20 RILs with two pure-culture reference strains of A. euteiches described in[41], i.e., the French RB84 strain from pathotype I andthe American Ae109 strain from pathotype III. CC re-sistance tests were also conducted in 2013 on all thelines with the RB84 strain only. The experiments used amodified version of the standardized test developed byMoussart et al. [54], as described in [41]. The pea lineswere evaluated for resistance to each strain, using a ran-domised complete block design with four and threeblocks in 2009 and 2013 tests, respectively, and fiveseeds per block sown in a pot. Resistance tests were con-ducted in a growth chamber (25 °C for 16h-day, 23 °Cfor 8h-night), as described in [41]. Seven-day old seed-lings were inoculated with a solution of 103 zoosporesper plant, produced as described by Moussart et al. [54].Disease severity was assessed on each plant seven daysafter inoculation, using a 0 (no symptoms) to 5 (deadplant) scoring scale proportional to the percentage ofbrowning symptoms on roots and epicotyls [41]. A CCroot rot index (CC_RRI) was calculated as the mean dis-ease score on all plants in a pot.In the field, the collection was evaluated for Aphano-

myces resistance in infested nurseries over four years(from 2010 to 2013) and three locations in France (Riec-sur-Belon, Finistère (RI); Dijon-Epoisses, Côtes d’Or (DI)described in [41]) and in the USA (Kendrick, Idaho(KEN)). The collection was phenotyped each year at thetwo French locations and in 2012 at KEN, USA, for atotal of nine environments. Field assays were carried outusing randomised complete block designs with threereplicates and an adjacent susceptible control (Solara)every two lines, as described in [41]. Two disease criteriawere used to assess resistance in each plot, as reportedin [41]: (i) a field root rot index (Field_RRI), using the0–5 CC scoring scale, evaluated on ten plants per plotfor each year in French nurseries and (ii) a field aerialdecline index (Field_ADI), measuring yellowing symp-toms on a plot, evaluated once or twice (Field_ADI1 and

Field_ADI2) in all the USA and French disease nurseries,using a 1 (green plant) to 5 or 8 (dead plant) diseasescoring scale depending on the nursery. The RRI andADI scores on each plot were adjusted for local diseasevariation measured by scores on the adjacent susceptiblecontrol, as described in [41].In the field, the collection was also phenotyped for de-

velopmental traits in a healthy nursery at Rennes-LeRheu (Ille-et-vilaine, France (REN)) for two years (2010and 2012), using a randomised complete block designwith three replicates, as described in [43]. Earliness wasscored on each plot as the number of calendar days to50 % bloom (FLO1), to 100 % bloom (FLO2) and to rip-ening defined as 100 % of dried plants (RIPE) from thefirst day of the year. Average plant height (HT) was eval-uated on five plants per plot at 100 % bloom.

GenotypingThe collection was genotyped using three morphologicalgenes (Af, afila/normal leaves; R, smooth/wrinkled seeds;A, anthocyanin production related to white/colouredflowers), specific-PCR markers for two known-functiongenes (Clpser and SugTrans, [55]), 45 simple sequencerepeat (SSR) markers from [56] and 13,204 SNP markersfrom [49]. DNA was extracted from each pea line fromapproximately 1 g of young leaves collected on plantsgrown in a greenhouse, as described by Doyle and Doyle[57]. Concentrations were adjusted to 10 ng/μl for SSRgenotyping and to 50 ng/μl for SNP genotyping.SSR fragments were amplified by polymerase chain re-

action (PCR) and analysed using GeneMapper® softwarev.4.0 (Applied Biosystems®, USA), as described in [44], ex-cept for the 20 RILs of the collection which were geno-typed for SSRs as in [41]. Out of the 45 SSRs, 28 and 17were located within and outside the main genomic regionspreviously associated to Aphanomyces resistance, respect-ively (Additional file 2) [41, 43]. Each SSR revealed be-tween two and 16 alleles in the collection. Each line of thecollection was coded for each SSR allele as homozygousfor the considered allele (“BB”), heterozygous (“AB”) orhomozygous for another allele (“AA”). A total of 441 SSRcoded alleles were included in the genotyping matrix.The 13,204 SNPs were part of a large pea SNP resource

of 248,617 robust filtered SNPs discovered from whole-genome Illumina HiSeq2000 sequencing of 16 diverse peagenotypes [49]. These 13,204 SNPs were all located ingene-context sequences, each originating from a separatetranscript [58], and were included in the Illumina Infinium®BeadChip developed by Tayeh et al. [49]. Infinium genotyp-ing and data analysis were conducted as described in Tayehet al. [49]. A total of 12,067 SNPs were selected for clearlybeing bi-allelic in the collection. Each line was coded “AA”or “BB” when homozygous for the first or second allele and“AB” when heterozygous.

Desgroux et al. BMC Genomics (2016) 17:124 Page 4 of 21

Statistical analysis of phenotypic dataPhenotypic data obtained on the collection for resistanceto A. euteiches, earliness and height were analysed, foreach variable in each field environment or CC experi-ment (individual analysis), and then for all year x loca-tion field environments (global analysis), using the R3.1.1 program [59]. In the individual analysis, phenotypicvariables were analysed using a linear model (LM) [Rfunction lm] including G (genotype) and R (replicate) asfixed factors. In the global analyses, E (environment) andGxE (genotype x environment) interaction were addedas fixed factors. Normality of residuals and homogeneityof their variances were checked using Skewness, Kur-tosis and Shapiro-Wilk statistics (P ≥ 0.05), as well asBartlett test (P ≥ 0.05), respectively [R functions skew-ness and kurtosis of package agricolae, [60]; plotresid ofpackage RVAideMemoire, [61]; bartlett.test]. Mean-based heritability (h2) was calculated for each variablefrom variance estimates in the individual and global LManalysis, using the formulas h2 = σG

2 /[σG2 + (σE

2/r)] and h2

= σG2 /[σG

2 + (σGE2 /E) + (σE

2/rE)], respectively, where σG2 is

the genetic variance, σGE2 the GxE interaction variance,

σe2 the residual variance, E the number of environments

and r the number of replicates per line. Least SquareMeans (LSMeans) were calculated from each LM ana-lysis (R function lsmeans of package lsmeans [62]). His-tograms of LSMeans frequency distributions were drawnusing the R function hist.Pearson correlation analysis was carried out between

LSMeans obtained from the individual and global ana-lysis (R function corr.test of package psych, [63]). Thesignificance of the Pearson correlations was evaluatedwith a false discovery rate correction for multiple testing(corrected p-value < 0.05; [64]). The heatmap of thePearson coefficients (r) was drawn using the R functionheatmap.2 (package gplots) [65].A Multiple factor analysis (MFA) was performed for

the different resistance variable categories (CC_RRI;Field_RRI; Field_ADI1 and Field_ADI2), usingLSMeans from individual and global LM analysis (Rfunction MFA of package FactoMineR, [66]). ForMFA, LSMeans missing values were handled with theR package missMDA [67].

Genetic analysisThe genotyping dataset of the collection, composed ofthree morphological genes, two specific-PCR markersfor known-function genes, 441 SSR alleles and 12,067SNP markers was filtered using PLINK 1.9 software [68–70]. Six of the 175 pea lines with missing data for morethan 10 % of SNP markers were not included in theGWA analysis. Markers with missing data that exceeded10 % or with a minor allele frequency (MAF) that didnot exceed 5 % in the 169 remaining lines, respectively,

were also removed for the GWA analysis. A total of9980 markers, including three morphological genes, twospecific-PCR markers of known-function genes, 189 SSRalleles and 9786 SNP markers were thus retained in thegenetic analysis.

Missing data imputationThe raw dataset of 9786 SNP markers was imputedusing the R function knncatimputeLarge (packagescrime, [71]). This function imputed missing values,which corresponded to 0.45 % of the total dataset, re-garding the k nearest neighbour SNPs without missingvalues. Imputation parameters were tested with 10 repli-cates, using a subset of 5001 SNP of the dataset with nomissing values. For each replicate, 0.45 % of missingvalues were randomly simulated. Parameters tested wereone to 10 nearest neighbours and four different methodsto determine distances between SNPs [72]. Distance cal-culation methods were based on corrected Pearson’scontingency coefficient, scaled Manhattan distance, sim-ple matching coefficient or Cohen’s kappa. Imputedvalues of simulated missing data were then compared toreal values. The lowest error rate mean (9.86 %) over the10 replicates was found with the Cohen’s kappa distancecalculation method and six nearest neighbours. Theseparameters were applied to the 9786-SNP-markers and189-SSR-allele dataset to impute missing values.

Linkage disequilibrium analysisPairwise LD between markers was explored within LGsfrom imputed genotypic data using PLINK 1.9 software.Obtained square correlation coefficient (r2) values werethen plotted against genetic distances (cM) according tothe consensus map from Tayeh et al. [49], namely TMapin this study, to estimate the LD decay. LD decay regres-sion curves were fitted to the observed LD decay plots(R 3.1.1 program; [59]), following Sved [73] method withrexpected2 = 1/(1 + 4Ne × c), Ne effective population sizeand c recombination rate between two markers. A non-linear model was fitted on the pairwise LD data, thenleast-squares estimates were computed (R function nls)and Ne was predicted from this fitted nonlinear model(R function predict). The LD decay rate of the popula-tion was measured as the genetic distance (cM) wherethe average r2 dropped to half its maximum value (r2 =0.5) [74].

Population structure and individual relatednessA principal Component Analysis (PCA) and a Kinshiprelatedness matrix were used to estimate the structure ofthe collection from genotypic data, using the EMMA(efficient mixed-model association) method in theGAPIT (Genome Association and Prediction IntegratedTool) R package ([75], see the GAPIT R script at [76]).

Desgroux et al. BMC Genomics (2016) 17:124 Page 5 of 21

PCA and Kinship matrices were calculated based on asubset of 2937 SNP markers, among the 9980 imputedfiltered markers on the TMap, corresponding to a singlerandomly-chosen marker per genetic position. Regardingto eigenvalues, the three first principal components(PCs) of the PCA were selected to take into accountstructure population in the GWA analysis. Clustering ofindividuals was considered based on their Kinship coor-dinates and was implemented in the GAPIT packagewith the UPGMA (Unweighted Pair Group Method withArithmetic mean) method.

Association mappingGWA analyses were performed using a modified versionof the multi-locus mixed model (MLMM) R package[11]. Briefly, the PCA matrix of population structure andthe Kinship matrix obtained from GAPIT were definedas cofactors in the MLMM (see the mlmm_cof.r R scriptat [77]). Significant SNP markers were also used as co-factors in a forward/backward approach.The initial script of mlmm_cof was modified to define

the multiple-Bonferroni (mBonf) threshold using theformula: mBonf = [−log(α/m)], with α = 0.10, the overallfalse positive threshold and m = 2937, the number ofmarkers selected at non-redundant genetic positions onthe TMap. Thus, the mBonf threshold was set at 4.47,which corresponded to a p-value of 2.5E-5. GWA studywas performed from LSMeans scores of all the variablesdescribed in the phenotypic data analyses section, as wellas from coordinates of lines on the two first PCs of MFAresistance variables analysis namely MFA.Dim.1 andMFA.Dim.2. In each GWA analysis, the optimal MLMMstep was determined as the largest stepwise mixed modelregression in which all cofactors have –log (p-value)above the mBonf threshold defined. At the optimalMLMM step, reliability of the analysis was evaluatedbased on plot of mBonf criteria among forward andbackward steps, qqplot at the optimal step with mBonfcriteria, and plot of partition of variance among steps(Additional file 3). The MLMM outputs used in thisstudy were the partition of variance (percentage of vari-ance explained by PCA, Kinship, markers in cofactors,and unexplained variance), the number and names ofmarkers as cofactors at the optimal step and the p-valueand allelic effect of each significant marker.Local LD analysis was used to define the CIs around

significant associated markers detected by GWA studyusing Plink 1.9 software. A LD block was determined asthe interval containing all markers in LD (r2 > 0.2) withthe significant associated marker, as described in [17].LD blocks were named with the LG name (Roman nu-meral) and then an Arabic numeral in genetic positionorder. A LD block was considered as consistent for eachtrait (resistance to A. euteiches, or earliness or height),

when it contained either one marker associated with atleast two variables of the trait, or two or more linkedmarkers (r2 > 0.2), each associated with at least one vari-able of the trait.

Haplotype analysisAt each consistent LD block associated with Aphano-myces resistance, marker haplotypes, i.e., genotypic pat-terns at the given LD block, were identified among allthe lines of the collection based on non-imputed rawgenotyping data. Haplotypes were named with the LDblock name and an Arabic numeral. For each trait sig-nificantly associated with marker(s) in a given consistentLD block, mean phenotypic scores of pea line groupscomprising more than eight lines (5 % of the total num-ber of lines) and carrying different haplotypes were com-pared, using the Tuckey-HSD test (α = 5 %; R packagemultcomp, [78]). Favourable and unfavourable haplo-types were defined as those meeting the three followingcriteria: (i) carrying favourable and unfavourable allele(s),respectively, at the disease trait-associated marker(s), (ii)without missing or heterozygous genotypic data at theother markers in the LD block and (iii) showing a sig-nificantly lower or higher disease mean score (breakingof group means considered, P < 0.05) than the otherfavourable or unfavourable haplotypes, respectively, forthe highest number of disease traits among those associ-ated with the LD block. According to these criteria,more than one favourable or unfavourable haplotype perLD block could be defined. Missing haplotypes were de-fined in lines showing missing genotypic data or hetero-zygosity for at least one marker in the considered LDblock. Each line of the collection was described for itsnumber of favourable haplotypes at all the consistent LDblocks. A Tukey-HSD test (α = 5 %) was then used tocompare the mean numbers of favourable haplotypes inthree groups of pea lines without missing haplotypes.The three groups were defined based on their MFA.-Dim1 scores and corresponded to lines with high (25 %lowest scores), intermediate (50 % interquartile scores)or low (25 % highest scores) levels of resistance,respectively.

Comparative mappingThe consensus map from Hamon et al. [43], namelyHMap in this study, summarized individual- and meta-QTL previously mapped for Aphanomyces resistanceand developmental traits [41, 43], while the consensusTMap of Tayeh et al. [49] contained all the markers usedin the present study. Thus, for comparative mapping,markers and QTL of the HMap were projected onto theTMap, using Biomercator V4.2 software [79]. The levelof connectivity between the two maps was estimatedusing the “InfoMap” tool of the software. Maps were

Desgroux et al. BMC Genomics (2016) 17:124 Page 6 of 21

compiled using the “iterative map projection” tool fromBiomercator V4.2 based on loci position data. TMap wasused as reference map and HMap was projected basedon common loci. Inversions of common loci were auto-matically resolved. Visualization of loci detected by asso-ciation and linkage mapping on the resulted consensusmap (namely THMap) was computed using MapChart2.1 software [80].

Identification of putative candidate genesUsing annotation data for SNP anchored sequences from[49], a search was carried out for the putative gene andprotein functions contained in each LD block for resist-ance to A. euteiches, earliness and height. The annota-tion data described predicted protein functions for eachtranscript sequence. The annotation was obtained fol-lowing Blastx searches against P. sativum, M. truncatula,G. max and A. thaliana protein sequences. If at leasttwo annotations were consistent, others missing, and atleast one e-value was lower than 1E-50, the correspond-ing putative protein function was assigned to the se-quence underlying the SNP. The putative annotation ofthe SNP-anchored sequence was scored with a ‘disagree-ing hits’ comment when annotations disagreed, and witha ‘not assigned’ comment when at least three annota-tions were missing and/or all e-values were higher than1E-50.. When possible, putative protein functions wereattributed to known mechanisms of plant developmentand responses to biotic stress, based on the literature.SNPs which were anchored to previously cloned peagenes and were located close to LD blocks associatedwith resistance or developmental variables were identi-fied based on information provided in [49].

ResultsAnalysis of phenotypic dataResistance to A. euteichesGlobal statistical analysis of RRI and ADI disease scores,obtained on the pea-Aphanomyces collection in fieldinfested nurseries over the nine environments studied,showed highly significant GxE interactions (P < 0.001)(Additional file 4). These results confirmed the relevanceof data analysis in each environment. Individual analysisof disease scores in each field environment, as well as ineach CC experiment, revealed highly significant geno-typic effects (P < 0.001) for all the disease variables butfor Field_RRI at RI in 2013 (P < 0.01) and Field_ADI2 atRI in 2010 (P < 0.05) (Additional file 4). Heritability ofresistance ranged from 0.28 (Field_RRI, RI 2013) to 0.96(Field_ADI2, DI 2012), depending on the variable, andwas high for most of the resistance variables (h2 > 0.60,except for Field_RRI at RI in 2013 and Field_ADI at RIin 2010, 2011, 2013 and KEN in 2012). Heritabilityvalues were especially high for CC_RRI scorings (h2 >

0.78), especially with the Ae109 strain (h2 = 0.91). Fre-quency distributions of LSMeans values for each individ-ual resistance variable tended to fit normal curves(Additional file 5), except for CC_RRI_09_Ae109 whichshowed a skewed distribution with some highly resistantlines.

Earliness and heightGlobal statistical analysis of earliness and height scores,obtained on the collection in the field healthy nursery atREN over two years (2010, 2012), showed highly signifi-cant GxE interactions (P < 0.001) (Additional file 4). In-dividual analysis of the scores in each year revealed ahighly significant genotypic effect (P < 0.001). Heritabil-ities of earliness and height traits in each environmentwere very high (h2 > 0.87). Frequency distributions ofLSMeans values for each individual developmental vari-able tended to fit normal curves and were consistent be-tween the two years (Additional file 5).

Correlations among variablesAll Field_RRI and CC_RRI scoring data were signifi-cantly and positively correlated between each other (cor-rected P < 0.001, 0.30 < r < 0.79), as well as closelyclustered, except that obtained with the Ae109 strain(Additional file 6). Most of the Field_ADI data wereslightly and positively correlated to CC_RRI data (cor-rected P < 0.01, 0.25 < r < 0.57) with the RB84 strain butnot with the Ae109 strain. Most of the Field_ADI1 andField_ADI2 scoring data (68 %) were also significantlyand positively correlated between each other (correctedP < 0.01, 0.22 < r < 0.79). Field_ADI scoring data weresignificantly and positively correlated with Field_RRIscoring data (corrected P < 0.03, 0.19 < r < 0.79), exceptthose from 2011 (RI and DI).All earliness and height data were significantly and

positively correlated between each other (corrected P <0.005, 0.26 < r < 0.92) (Additional file 6). Earliness datawere slightly and negatively correlated with most of theField_ADI data (corrected P < 0.05, −0.5 < r < −0.2).

Multiple factors analysisThe two first PCs of the MFA analysis of disease resist-ance variables explained a total of 56 % of the inertia(MFA.Dim.1: 44.13 % and MFA.Dim.2: 11.89 %; Fig. 1a).Three groups of variables could be distinguished, includ-ing CC_RRI, Field_RRI and Field_ADI. A total of 77 % ofthe variables, especially the CC_RRI_RB84 variables, werehighly correlated with MFA.Dim.1 (r2 > 0.5) and well rep-resented on the first axis (cos2 > 0.5) (Additional file 7).Pea lines with extreme negative or positive coordinates

on MFA.Dim.1 were considered as the most resistant orsusceptible lines, respectively. MFA.Dim.2 was mainlyconstructed from RB84 CC_RRI variables (r2 ≤ −0.5) in

Desgroux et al. BMC Genomics (2016) 17:124 Page 7 of 21

contrast to Field_ADI variables (r2 > 0.5). Pea lines withextreme negative or positive coordinates on MFA.Dim.2had low levels of aerial symptoms or root symptoms, re-spectively. Dispersion of the pea lines on the two princi-pal MFA components showed that the frequency ofresistant and susceptible lines was homogeneous amongthe collection and within some groups of lines (Fig. 1b),which is optimal for GWA analysis of the resistance.The three most resistant lines were GSP breeding lines,namely AeD99OSW −45-8-7 (A100), AeD99OSW −50-2-5 (A104) and AeD99OSW −47-6-1 (A101). The two mostsusceptible lines were Safranos (A154) and Marignan(A153), registered as French varieties.

Genetic analysesLinkage disequilibrium, structure, and kinship in thepea-Aphanomyces collectionFrom the imputed genotypic data of the collection ob-tained with the filtered 9980 markers, the LD decay wasestimated to range from 0.08 to 0.16 cM, depending onthe LG, and averaged 0.12 cM over all the pea LGs(Additional file 8). Based on non-redundant marker po-sitions on the consensus genetic map of Tayeh et al. [49](TMap), the average distance between two markers usedin the analysis was 0.27 cM.From the defined subset of 2937 SNPs located at non-

redundant genetic positions on the TMap, PCA analysisof the collection identified three first PCs that explained

a total of 20.91 % of genetic variation in the collection(Additional file 9 D). The first PC contributed to 9.28 %of the variation and the second and third PCs to 6.34and 5.29 %, respectively. Pea lines were clustered slightlyon PCs 1 to 3 depending on their origin, mainly var-ieties, distinct RIL populations, and groups of GSPbreeding lines with any, two or three partially resistantparents in their pedigree (Additional file 9 A to C). Thefirst three PCs were added to the GWA model as cofac-tors to take into account this slight population structure.From the subset of 2937 SNPs, the Kinship matrix ofgenetic similarities revealed a moderate relatednessamong lines of the collection (0.5 < r2 < 0.65 for themajor part of the lines), including sub-groups withhigher relatedness (r2 > 0.8) according to pedigree, enduse or sowing type (Additional file 10). Thus, the Kin-ship relatedness matrix was also added to the GWAmodel as a cofactor.

GWA markers and confidence intervals

Disease resistance GWA analysis identified a total of79 markers, located on the seven LGs, associated with33 global or individual Aphanomyces resistance vari-ables, including (i) 28 Field_RRI or _ADI variables fromnine field environments, (ii) the three CC_RRI variablesand (iii) the two MFA variables (Table 1 and Additionalfile 11). Zero to nine markers were significantly

A B

MostsusceptiblelinesMost

resistantlines

Low RRI

Low ADI

A100A101

A104

A154

A153

Fig. 1 Graphical representation of Multiple Factors Analysis (MFA) of the phenotypic data. Correlation circle of variables (coloured according togroups) (a) and genotype factor map (coloured according to origin) (b), on the first two principal components of the MFA of disease variables.Earliness and height variables are supplemental variables and thus are projected onto the dimensions but were not included in the analysis. Allthe variables are abbreviated as described in Table 1. GSP0-3: lines from the GSP breeding program containing in their pedigree zero to three ofthe previously described major sources of resistance (PI180693, 90–213 and 552) [43] (Additional file 1), respectively. Parents: Parental lines of GSPbreeding programs and/or RIL populations. BRILs iia, iib, iic: Selected lines from the Baccara x PI180693, Baccara x 552 and DSP x 90–2131 RILpopulation, respectively; New sources: Selected accessions from the large INRA screening program; Varieties: Registered dry pea varieties. Extremegenotypes are labelled on the two main principal components

Desgroux et al. BMC Genomics (2016) 17:124 Page 8 of 21

Table 1 Genome-wide association analysis results using the multi-locus mixed model (MLMM) method in the pea-Aphanomycescollection

Variable (a) Number ofmarkers (b)

Range ofp-value (c)

Range of alleliceffect (d)

% of phenotypic variance explained by Unexplainedvariance (h)PCA (e) Kinship (f) Markers (g)

Field_RRI_All 2 7.86E-11–8.10E-06 0.13–0.17 48 % 20 % 19 % 14 %

Field_RRI_10_RI 2 1.81E-05–2.37E-05 0.16–0.19 13 % 28 % 25 % 34 %

Field_RRI_11_RI 2 9.92E-08–6.58E-06 0.23–0.26 42 % 9 % 20 % 29 %

Field_RRI_12_RI 5 3.14E-07–1.50E-05 0.17–0.29 20 % 18 % 39 % 23 %

Field_RRI_13_RI 0 – – 26 % 25 % 0 % 48 %

Field_RRI_10_DI 1 5.78E-06 0.17 24 % 28 % 10 % 38 %

Field_RRI_11_DI 1 5.90E-07 0.29 54 % 25 % 9 % 12 %

Field_RRI_12_DI 3 4.76E-06–7.46E-06 0.15–0.17 36 % 14 % 23 % 28 %

Field_RRI_13_DI 3 6.00E-08–2.17E-05 0.16–0.26 28 % 13 % 23 % 37 %

CC_RRI_09_RB84 1 5.40E-12 0.28 43 % 8 % 19 % 30 %

CC_RRI_09_Ae109 2 8.82E-28–1.60E-06 0.27–0.81 8 % 10 % 68 % 13 %

CC_RRI_13_RB84 3 7.34E-11–3.89E-06 0.17–0.26 47 % 6 % 22 % 25 %

Field_ADI1_All 4 7.86E-11–2.43E-06 0.20–0.32 32 % 25 % 27 % 16 %

Field_ADI1_10_RI 4 3.12E-08–2.25E-05 0.26–0.40 10 % 4 % 39 % 48 %

Field_ADI1_11_RI 0 – – 5 % 50 % 0 % 45 %

Field_ADI1_12_RI 3 3.99E-07–8.98E-06 0.45–0.54 23 % 50 % 15 % 13 %

Field_ADI1_13_RI 5 3.85E-08–3.32E-05 0.13–0.25 31 % 0.001 % 37 % 32 %

Field_ADI1_10_DI 0 – – 37 % 45 % 0 % 18 %

Field_ADI1_11_DI 0 – – 2 % 17 % 0 % 81 %

Field_ADI1_12_DI 7 1.13E-10–2.53E-05 0.21–0.61 29 % 21 % 39 % 12 %

Field_ADI1_13_DI 2 2.53E-10–1.16E-08 0.36–0.43 35 % 15 % 26 % 23 %

Field_ADI1_12_KEN 1 2.64E-05 0.18 32 % 25 % 8 % 34 %

Field_ADI2_All 3 1.15E-07–8.27E-06 0.22–0.38 39 % 47 % 8 % 6 %

Field_ADI2_10_RI 0 – – 15 % 0.004 % 0 % 85 %

Field_ADI2_11_RI 2 7.77E-07–3.61E-06 0.33–0.43 0 % 43 % 23 % 34 %

Field_ADI2_12_RI 0 – – 39 % 47 % 0 % 14 %

Field_ADI2_13_RI 0 – – 39 % 45 % 0 % 17 %

Field_ADI2_10_DI 6 7.56E-08–1.27E-05 0.26–0.46 20 % 0.002 % 38 % 42 %

Field_ADI2_12_DI 0 – – 24 % 50 % 0 % 26 %

Field_ADI2_13_DI 9 3.25E-12–2.34E-5 0.26–0.45 36 % 5 % 45 % 14 %

Field_ADI2_12_KEN 1 2.55E-05 0.34 9 % 48 % 11 % 32 %

MFA.Dim.1 4 1.51E-08–1.26E-05 0.42–0.56 56 % 15 % 21 % 8 %

MFA.Dim.2 2 1.01E-05–1.48E-05 0.36–0.48 9 % 78 % 1 % 13 %

FLO1_10_REN 5 1.24E-17–1.97E-05 1.82–5.86 9 % 17 % 64 % 10 %

FLO1_12_REN 7 6.29E-16–3.31E-06 1.55–12.80 8 % 23 % 63 % 6 %

FLO2_10_REN 8 8.50E-12–2.21E-06 0.91–2.88 5 % 0.001 % 71 % 24 %

FLO2_12_REN 3 2.68E-15–3.69E-08 2.47–5.49 5 % 19 % 52 % 24 %

RIPE_10_REN 7 5.02E-13–7.74E-06 0.94–2.88 0 % 3 % 72 % 25 %

Desgroux et al. BMC Genomics (2016) 17:124 Page 9 of 21

associated with each variable, with a p-value that rangedfrom 8.82E-28 to 3.32E-05, depending on the marker.Zero to nine cofactors were thus retained in the MLMMfor each disease resistance variable, explaining a total ofzero to 68 % of the phenotypic variation depending onthe variable and in accordance with the heritability ofthe trait. Missing heritability (unexplained variance)ranged from 6 to 85 % and PCA and Kinship individu-ally captured between zero and 78 % of the variance, de-pending on the variable. Allelic effects of markersassociated with Field_ADI variables fluctuated widely,depending on the marker (absolute values from 0.13 to0.61) (Table 1 and Additional file 11). Markers associ-ated with Field_RRI had similar allelic effects as thoseassociated with CC_ RRI with the RB84 strain (absolutevalues from 0.13 to 0.29), but lower than those associ-ated with CC_RRI with the Ae109 strain (absolute valuesfrom 0.27 to 0.81).A total of 52 CIs, ranging from 0 to 5.4 cM, were de-

fined around all the significant disease-trait-associatedmarkers, which included markers in LD (r2 > 0.2) withthe targeted marker(s). Three kinds of LD blocks wereidentified, based on their significance and consistency(Fig. 2): (i) Fourteen LD blocks were considered as con-sistent since these included two to six disease trait-associated markers (double red stars on Fig. 2; 3.25E-12< P < 2.64E-5). (ii) Four disease LD blocks included asingle highly significant disease trait-associated marker(single red stars on Fig. 2; 8.82E-28 < P < 1.13E-10).Three of these markers were associated with CC_RRIvariables. (iii) Most of the disease LD blocks (67 %) in-cluded a single moderately significant disease-trait-associated marker (3E-08 < P < 3.3E-05).Out of the 14 consistent LD blocks, seven were Fiel-

d_ADI specific, three were Field_RRI specific and fourwere common to Field_ADI and _RRI. The sevenmarkers associated with global Field_ADI variables wereall located in common LD blocks as individual Fiel-d_ADI variables. Out of the two markers associated tothe global Field_RRI variable, one was located in thesame LD block as one containing individual Field_RRIvariables. Even if CC_RRI and Field_RRI data were

correlated (r2 > 0.5), no common LD block was detected.Among the four markers associated with MFA.Dim.1,two were included in Field_RRI specific LD blocks andone in a LD block that contained Field_ADI associatedmarkers. The two markers associated with MFA.Dim.2variable were included in Field_ADI specific LD blocks.The THMap, which resulted from the projection of

the HMap onto the TMap, was based on 144 commonmarkers (13 to 29 markers per LG; Additional file 12),most of which were located in previous QTL regions.The projection led to a total of 41 inversions of markerpositions, which did not likely affect the QTL projection.Comparison of linkage and association mapping re-

sults revealed that 19 of the 27 genomic regions includ-ing individual Ae-Ps QTL and 12 of the 27 MQTL-Aemeta-QTL previously reported [41, 43] were also de-tected by association mapping (Fig. 2 and Additional file11). (i) Eight out of the 14 consistent disease LD blocks(I.5, II.1, III.3, III.4, III.6, III.9, VII.13 and VII.14) co-localized with four of the seven main Ae-Ps QTLs previ-ously identified and used for NILs construction byLavaud et al. [44], i.e., Ae-Ps1.2, Ae-Ps2.2, Ae-Ps3.1a-band Ae-Ps7.6a-b. Moreover, the highly consistent LDblock VII.16 was located just 0.45 cM away from Ae-Ps7.6b. Two other consistent LD blocks co-localizedwith two less consistent Ae-Ps QTL (Ae-Ps3.2 and Ae-Ps7.3). (ii) One of the four disease LD blocks includingsingle highly significant disease-trait-associated markers(IV.8) co-localized with the main previously detectedQTL Ae-Ps4.5. (iii) Among the 40 remaining disease LDblocks identified in this study, three co-localized withthe previous main QTL Ae-Ps4.1 and 11 were located inregions not previously reported for resistance to A.euteiches. The main QTL Ae-Ps5.1 previously identifiedon LGV was not detected in this study.

Linkage between resistance and morphological or de-velopmental traits A total of 25 markers, distributedover four LGs, were significantly associated (1.24E-17 <P < 1.97E-05) with the six earliness variables scored inthe pea-Aphanomyces collection in the REN healthy

Table 1 Genome-wide association analysis results using the multi-locus mixed model (MLMM) method in the pea-Aphanomycescollection (Continued)

RIPE_12_REN 2 2.69E-17–4.63E-06 2.15–3.39 0 % 25 % 49 % 26 %

HT_10_REN 3 3.76E-26–5.00E-06 6.42–30.65 0 % 37 % 53 % 10 %

HT_12_REN 2 1.62E-23–2.55E-05 8.10–34.61 0 % 23 % 76 % 1 %

(a) Variables are abbreviated as follows: CC Controlled conditions experiments, Field Infested field experiments, RRI Root rot index, IDA1 First aerial decline index,IDA2 Second aerial decline index, FLO1 date of 50 % bloom, FLO2 date of 100 % bloom, RIPE date of 100 % dried plants, HT height of plants, 09 to 13 year of fieldevaluation, All Global variables over field environments, RI Riec-sur-Belon, France, DI Dijon-Epoisses, France, REN Rennes-Le Rheu, France, KEN Kendrick (ID), USA,RB84 and Ae109 two pure-cultured strains; (b) number of markers used as cofactors at the optimal step of the multi-locus mixed model (MLMM) analysis; (c) Rangeof p-values of the significant markers, significance threshold is p-value < 3.4E-05 as described in Methods section; (d) Range of allelic effects of the significantmarkers in absolute values; Percentage of phenotypic variance explained by: (e) the principal component analysis (PCA) matrix of the collection, (f) the Kinshiprelatedness matrix among lines of the collection, (g) all cofactor markers and (h) the unexplained variance qualified as “missing heritability”

Desgroux et al. BMC Genomics (2016) 17:124 Page 10 of 21

field nursery over two years (Table 1 and Additional file11). Twenty LD blocks were defined around earliness-associated markers, covering 0.3 to 5.4 cM. Six wereconsistent, since they were associated with two to sixvariables, and had different allelic effects ranging fromone to 8.4 days (double blue stars on Fig. 2; 1.24E-17 <P < 1.97E-05). Three LD blocks were not consistent butincluded highly significant single trait-associatedmarkers and had high allelic effects from 2.5 to 12.8 days(single blue stars on Fig. 2; 6.29E-16 < P < 4.30E-10).

Three of the 14 QTL previously identified in Hamon et al.(2013) for earliness, i.e., Flo-Ps2.2, Flo-Ps2.3 and Flo-Ps3.1,were also detected by association mapping.A total of five markers, corresponding to three LD

blocks distributed over two LGs, were significantly asso-ciated with the two height variables scored (3.76E-26 <P < 2.55E-05). Two LD blocks (III.7 and VII.3) on LGIIIand LGVII, ranging from 0.5 to 2.4 cM, were consistent,since they each included two height-associated markers.In particular, LD block III.17 was highly significant

I.5 **

I.1

I.2

I.3

I.4

LGI(113 cM)

II.1**

LGII(118 cM)

II.2 *

II.3 **

II.5 *

II.6

II.7

II.4

III.1

III.2 ** *

III.3**III.4**III.5

III.6**III.7

III.8III.9**

III.10

III.12III.13

III.14

III.15**

III.16

III.17 **

III.18

LGIII(136 cM)

III.11

LGIV(121 cM)

IV.1IV.2

IV.3

IV.4

IV.5IV.6IV.7

IV.8 *

IV.9

IV.10

IV.11**

IV.12**

IV.13

VI.5

VI.4VI.3

VI.2

VI.1

LGVI(115 cM)

LGV(117 cM)

V.1

V.2

V.3

LGVII(123 cM)

VII.1VII.2

VII.3** **

VII.4VII.5 **VII.6

VII.7 *VII.8 *VII.9VII.10VII.11

VII.12VII.13** **VII.14**VII.15

VII.16** **

VII.17 *

VII.18** **

Legend

Linkage mapping QTL

MQTL-Ae[43]

Ae-Ps QTL[41,43]

MQTL-Morpho[43]

Flo-Ps QTL[43]

Ht-Ps QTL[43]

Confidence intervals oftrait-associated markers

Resistance toA. euteiches

Earliness

Height

Global Field RRI

CC RRI Ae109

CC RRI RB84

Field RRI

Field ADI

FLO1

Height

Trait-associatedmarkers

MFA.Dim.1

Global Field ADI

MFA.Dim.2

FLO2

RIPE

Aphanomyces resistance

Earliness and height

Marker significance

P-value E-5

a b

Fig. 2 Comparative genetic map of genome-wide association (GWA) and previously detected linkage quantitative trait loci (QTL). The comparativegenetic map was constructed from the projection of the consensus map from Hamon et al. [43] onto the consensus map from Tayeh et al. [49].Linkage groups (LG) are named from I to VII and their size is indicated in cM Haldane. Resistance-, earliness- and height- associated markers, LDblocks and QTL are indicated in red, blue and black, respectively. To the right of each LG: Names of significant trait-associated markers identifiedby GWA study are indicated. The shading in the LG bar represents the confidence intervals around the significant trait-associated markers, basedon linkage disequilibrium (LD) value of r2 > 0.2. Symbols shown on the right of each marker and described in the legend indicate trait(s) to whichthe marker was significantly associated. Sign width is proportional to the significance level (p-value) of the marker-trait association. Names of LDblocks are indicated to the extreme right of each LG. The brackets indicate markers that were attributed to a same LD block (r2 > 0.2). LD blockslabelled by double and single asterisks correspond to consistent (at least two significant markers with P < 5E-10) and highly significant but notconsistent (one significant marker with P < 5E-10) blocks, respectively. Genomic positions of cloned pea genes are indicated in grey. The resistancegenes cluster identified by Tayeh et al. [49] is represented by light grey shading on the bottom of LGIII. To the left of each LG: Projected QTL andMeta-QTL described in [41, 43] are represented with the same colour trait codes as for markers detected by GWA study. Solid bars representMeta-QTLs [43], while hatched bars represent initial QTL before meta-analysis [41, 43]. The main Aphanomyces resistance QTL and Meta-QTL [43]names are in bold italic

Desgroux et al. BMC Genomics (2016) 17:124 Page 11 of 21

(3.76E-26 < P < 1.62E-23) and presented an allelic effectwhich contributed to height differences of more than 30cm. None of the three LD blocks co-localized with previ-ous QTL detected for height in [43] (HT-Ps).Out of the total 75 LD blocks identified for resistance

to A. euteiches, earliness and height, only five (II.1, III.2,VII.13, VII.16 and VII.18) were detected for both resist-ance and earliness, and one (VII.3) for both resistanceand height (Fig. 2 and Additional file 11). In the five re-sistance and earliness common LD blocks, three SNPmarkers and one SSR marker (AA387) were associatedwith both resistance and earliness. Allelic effects at thesemarkers were opposing for resistance and earliness, i.e.,the resistance-enhancing alleles conferred later bloom.One LD block identified for resistance to A. euteiches

also co-localized with the Af morphological gene whichcontrols leaf type on LGI. No LD block co-localized withthe A morphological gene (i.e., PsbHLH gene) whichcontrol anthocyanin production on LGII. Linkage be-tween Aphanomyces resistance and normal leaves at theAe-Ps1.2 QTL reported in [41, 43] was thus confirmed,while linkage between resistance and coloured flowers atthe Ae-Ps2.2 QTL was broken.

Marker haplotypesAt each of the 14 consistent disease LD blocks, three to26 haplotypes were identified, depending on the LDblock (Additional file 13). Mean comparison of pheno-typic LSMeans between marker haplotype groups of aLD block, allowed one or two favourable haplotypes perLD block to be selected, except at LD block IV.12 forwhich four favourable haplotypes were identified. A totalof 22 favourable haplotypes were identified over the 14consistent disease resistance LD blocks. A total of 112haplotypes, carrying at least one favourable allele at thedisease-trait-associated marker(s) of the blocks, werealso identified among the 14 consistent disease resist-ance LD blocks. In a subset of 84 lines without missinghaplotypes, the 25 % most resistant lines, according toMFA.Dim.1 coordinates, showed a mean frequency offavourable haplotypes which was significantly higher(eight favourable haplotypes on average for the 14 LDblocks) than that of the lines classified in the intermedi-ate or susceptible groups (six and four favourable haplo-types on average, respectively) (Fig. 3 and Additional file14). Lines AeD99OSW-49-5-7 (A103), AeD99OSW-45-8-7 (A100) and AeD99OSW-37-3-4 (A092) showed thehighest number of favourable haplotypes defined (13, 12and 11 respectively) at the 14 consistent disease LDblocks in the subset of 84 lines. The group of susceptiblelines carried the highest number of unfavourable haplo-types (three unfavourable haplotypes on average, versustwo and one for intermediate and resistant groups, re-spectively). Two of the favourable haplotypes for

resistance to A. euteiches (II.1.02 and VII.16.07) were theworst haplotypes for earliness as it gave the latest flower-ing time. Another one (II.1.06) was unfavourable forearliness as it gave intermediate flowering time. One ofthe favourable haplotypes for resistance to A. euteiches(VII.3.04) was unfavourable for plant height as it gavehigher plants. Early bloom and small height are two im-portant selection criteria in dry pea breeding programs.

Putative candidate genesOut of the 550 markers in the 52 disease-related LD blocks,211 were located in gene sequences and could be assignedto putative protein functions (Additional file 15). Out ofthese 211 annotated SNPs, 135 SNPs corresponded togenes with putative functions reported to be involved inplant response to biotic stresses. Among these genes, 21 en-code for protein domains reported as involved in plant rec-ognition of pathogens (including Leucine Rich Repeat-LRR-domains), 65 for domains reported to contribute to signaltransduction (including Serine-Threonine-Tyrosine ProteinKinases and proteins involved in Ethylene biosynthesis), 12for proteins playing a role in transcription regulation (in-cluding a WRKY transcription factor, a VQ motif and abasic helix-loop-helix domain) and 47 for proteins involvedin plant defence responses such as cell wall modifications(including homeobox leucine zipper proteins), oxidativeburst (including Ras GTPases), detoxification (including anABC transporter C and cytochrome P450 enzymes) or celldeath (including chaperone Dnaj proteins). A subset of 24putative proteins were also reported to be involved in plantgrowth or development, out of which 14 were not reportedto be related to pathogen resistance. Four disease-trait-associated markers in LD blocks III.14, IV.10,V.1 and VI.1,were close to (0.1–4.4 cM) but not in LD with pea resist-ance genes analogues (RGA) coding for nucleotide bindingsite Leucine Rich Repeat (NBS-LRR) proteins (RGA1.1,RGA2.75, RGA1.5 and RGA2.159, respectively) [81, 82].Three RRI-associated markers in LD blocks III.12, IV.8 andVII.11, and one Field_ADI-associated marker in LD blockVI.5, were close (0.5–0.8 cM) to cloned genes involved inpea-rhizobia symbiosis (SYM7, SYM9, SYM29 and SYM8,respectively). A Field_ADI- associated marker in LD blockI.5 co-localized with the SGR gene (Stay Green) which con-trols chlorophyll catabolism during foliage senescence [83].Among the 18 consistent or significant disease-related LDblocks, 56 SNP markers were attributed to putative se-quences coding for interesting candidate proteins involvedin plant resistance to pathogens, including Leucine-rich re-peats (LRR) in LD blocks I.5, IV.8 and VII.3, Serine/threo-nine protein kinases in six LD blocks [84] and an ABCtransporter [85] in LD block III.6.Among the 15 additional LD blocks exclusively related

to earliness, 55 of the 110 SNP markers could be assignedto a putative function, 24 of which are involved in plant

Desgroux et al. BMC Genomics (2016) 17:124 Page 12 of 21

growth or development. Three earliness-associatedmarkers in LD blocks II.3, III.2 and III.5, were close (0–2.1 cM) to cloned genes involved in flowering time or in-determinacy (PsFLD, Hr and PEAFLO respectively) [86–88]. Among the three height-related LD blocks, 26 of the44 SNP markers could be assigned to a function. Eight ofthem were reported to be involved in plant growth anddevelopment, including two SNPs in LD blocks III.11 andIII.17 designed in sequences of pea cloned genes control-ling plant height, i.e., La (PsLA) [89] and Le (PsGA3ox1)[90], respectively. The height-associated markers in thethird height-related LD block VII.3 was located just0.8 cM from the SNP marker designed in the sequence ofthe Cry gene, involved in plant growth [89].

DiscussionThis work describes the first GWA approach in pea todecipher polygenic control of partial resistance to A.euteiches, using novel plant material as well as highdensity and quality genomic resources. The study evalu-ated a pea-Aphanomyces collection of 175 lines,enriched in sources of resistance to A. euteiches, whichwere identified in the last 15-years of research andbreeding programs. Genetic diversity and recombinationevents screened in the collection were higher than thosepreviously studied in bi-parental populations. We tookadvantage of the GenoPea Infinium ® BeadChip recentlygenerated [49], containing 13,204 SNPs precisely an-chored onto a consensus genetic map generated from

Num

ber

of fa

vour

able

mar

ker

hapl

otyp

es

0

2

4

6

8

10

12

a

b

c

A

MFA.Dim.1 coordinates

Fre

quen

cy

−5 −4 −3 −2 −1 0 1 2 3 4 5

0

10

20

30

40

50 1st quartile 3rd quartileInterquartile

B

Fig. 3 Mean comparison of favourable haplotype numbers in resistant, intermediate and susceptible lines. a: Box plot of the number offavourable linkage disequilibrium (LD) block haplotypes carried by pea lines (without missing haplotypes) from the pea-Aphanomyces collectionbelonging to the first quartile (25 % lowest values), interquartile (50 % intermediate values) and third quartile (25 % highest values) of coordinateson the first principal component of the multiple factors analysis (MFA.Dim.1). The three groups include pea lines with a high (first quartile),intermediate (interquartile) and low (third quartile) level of resistance. Letters indicate significantly different groups based on mean comparison(Tuckey-HSD, α = 0.05). b: Distribution histogram of MFA.Dim.1 for the pea lines (without missing haplotypes) of the collection. The first quartile(25 % lowest coordinates; dark green) carries the highest number of favourable haplotypes, the interquartile (intermediate values; light green)carries an intermediate number of favourable haplotypes and the third quartile (25 % highest values; light red) carries the lowest number offavourable haplotypes

Desgroux et al. BMC Genomics (2016) 17:124 Page 13 of 21

genotyping data on more than 1300 RILs. This novelgenomic resource greatly enhanced the resolution of thegenetic analysis in the association mapping collectioncompared to the few hundred genetic markers used inprevious linkage studies of Aphanomyces resistance inpea [41, 43]. We used the same resistance phenotypingmethodologies and Aphanomyces field network as in theprevious QTL analysis of resistance, which led to highaccuracy when comparing the results of these two ap-proaches. Finally, this study used a MLMM to performGWA study. This model corrects for population struc-ture and individual relatedness, and takes into accountsignificantly associated SNPs as cofactors to detect otherSNPs. This approach limits the number of false positiveand false negative associations [11].Specifically, the GWA study (i) identified 14 consistent

loci out of a total of 52 small sized LD-based CIs de-tected for resistance to A. euteiches; (ii) validated and re-fined six of the seven major resistance QTL previouslyidentified; (iii) confirmed or broke negative associationswith morphological and developmental loci; (vi) pin-pointed interesting haplotypes and determined their fre-quency in the most resistance pea lines, for applicationin breeding; (v) identified relevant and interesting puta-tive candidate genes underlying main resistance loci.

GWA study validated most of the previously detectedQTL and identified new resistance lociThis study validated most of the previously detectedAphanomyces resistance QTL in pea. Out of the 52resistance-associated loci, 41 were estimated to be lo-cated in 19 of the 27 projected Ae-Ps QTL, including sixof the seven main consistent QTL, and 12 of the 26 pro-jected meta-QTL (MQTL-Ae) previously identified frombi-parental population studies [41, 43]. Four and two ofthe main previously consistent resistance QTL (Ae-Ps1.2, Ae-Ps2.2, Ae-Ps3.1, Ae-Ps7.6a-b and Ae-Ps4.1, Ae-Ps4.4-4.5) were re-detected in this study by eight of the14 consistent resistance loci and seven of the othersingle-variable specific loci, respectively. Twelve previ-ously detected minor-effect QTL were also confirmedwith 18 resistance-associated loci, including two thatwere consistently detected. GWA validation of previouslinkage resistance QTL was expected, as the pea-Aphanomyces collection was enriched in genetic poolsderived from sources of resistance studied in previousQTL analysis (90–2131, 552 and PI180693). A total of87 % of the 121 GSP breeding lines and RILs of the col-lection were derived from at least one of these threesources of resistance. In crops, GWA studies usuallyused panels with a good representation of species gen-etic diversity, such as diversity collections developed inpea [52, 91]. For pea Aphanomyces resistance, the GWAstudy required an adapted panel with a higher resistance

allele frequency than to the one previously observed inthe pea natural genetic variation (Pilet-Nayel et al., un-published) [36]. However, Kwon et al. [50] identified aSSR marker (PSARGDECA_F) associated to Aphano-myces resistance [34] in the USDA pea core-collection,which is estimated to be located in the Ae-Ps7.6 QTL re-gion, close to LD block VII.11, using comparative geneticmaps [43, 45, 49]. Marker and methodology tools werealso used to optimize comparative analysis of GWA andprevious linkage results. Most of the SSR markers, previ-ously associated with resistance QTL [43], were inte-grated into the consensus SNP genetic map of Tayeh etal. [49]. Common markers were used as bridges for com-parative mapping through linkage QTL projection. Fur-thermore, the field and climate chamber conditions werethe same in [41, 43] and in this study, which increasedthe accuracy of comparative mapping.Eight of the 27 projected Ae-Ps QTL previously associ-

ated with the resistance [43] were not identified byGWA in this study, including one main consistent QTL(Ae-Ps5.1). Several factors could lead to non-validationof QTL by GWA. These include the low detectionpower of minor-effect QTL, low allele frequency, GxEinteractions, epistatic interactions and gaps in markercoverage [5]. In this study, the detection power of loweffect QTL may have not been optimal due to the mod-erate population size (175 lines). Markers with low MAF(MAF < 5 %) were sometimes clustered in some genomicregions, such as on LGV, and were removed from theanalysis, creating gaps in the genome-wide scan ofmarker-trait associations. GxE interactions for fieldAphanomyces resistance were observed in this studyover the nine environments tested, as previously re-ported and discussed regarding pedo-climatic condi-tions, pathogen diversity and the presence of other rootrot pathogens [60]. They may have resulted in the detec-tion of QTL specific to the field environments presentlystudied compared to previous ones, which were submit-ted to different climatic conditions and pathogen popu-lation pressures. Epistatic interactions have also beenreported for Aphanomyces resistance [41] and someQTL effects were recently shown to increase or be lostwhen transferred into a new genetic background [44]. Fi-nally, despite the unprecedented marker density used inGWA analysis in pea, marker coverage gaps cannot beexcluded, due to the restricted choice of SNPs in theInfinium SNP chip used and the unavailable informationon physical distances between markers. The non-detection of the main consistent QTL Ae-Ps5.1 was par-ticularly unexpected since it was identified from CC ex-periments with several strains including RB84 andAe109 used in this study, and contributed up to 38 % ofthe phenotypic variation in the DSP x 90–2131 RILpopulation [43]. However, Lavaud et al. [44] recently

Desgroux et al. BMC Genomics (2016) 17:124 Page 14 of 21

showed that the effect of Ae-Ps5.1 was lost when trans-ferred into pea lines other than DSP, suggesting epistaticinteractions between this QTL and genetic background.In addition, a very low allele frequency was observed forthe SNP designed in the Ae-Ps5.1 closely linked markerPsLD (MAF = 0.8 %) [41, 86] and for its neighbour SNPin high LD (r2 = 0.55; MAF = 1.4 %), which were bothconsequently removed from the analysis.In this study 11 new Aphanomyces resistance loci

were identified, including three consistent ones (IV.11,IV.12 and VII.18). Comparative mapping was based onfew markers (mainly SSRs) which resulted in some in-versions. Thus, some projection inaccuracies may havewrongly pointed out these loci as new ones, especiallyfor the VII.16 interval. However, the higher resolution ofGWA analysis may also have highlighted resistance al-leles from the studied germplasm, which would not havebeen previously detected due to low marker coverage.Most probably, new favourable alleles are present in thepea-Aphanomyces collection, which included newsources of resistance [36] which were not investigated inprevious linkage-QTL analyses.In other pathosystems, GWA mapping often validated

linkage mapping QTL and detected new loci. For ex-ample, Samayoa et al. [15] and Zhao et al. [14] reportedsimilar results when comparing linkage and associationmapping of Mediterranean corn borer resistance inmaize and Sclerotinia sclerotinium resistance in soybean,respectively.

GWA study detected resistance loci with a higherresolution than in previous linkage mapping studiesIn the present GWA study, QTL resolution increasedcompared to previous linkage mapping studies of Apha-nomyces resistance. The 52 resistance-associated inter-vals were detected with much smaller sizes (0–5.2 cM,depending on the locus, 0.9 cM in average) than thosepreviously identified in QTL mapping studies (2.1–43.9 cM, depending on the Ae-Ps QTL, 17.4 cM in aver-age). The LD decay rate in the pea-Aphanomyces collec-tion was low (r2 < 0.5 within a map distance of 0.12 cM)compared to the one observed in a similar study in pea[9] (average r2 < 0.17 within a map distance of 5–10 cM).This low LD decay rate is particularly interesting as LDdecay of autogamous (self-pollinated) species have gen-erally been reported slow (~100kb), as in rice, foxtailmillet or soybean, compared to outcrossing species(~2kb) [5]. This may be especially due to a high numberof recombination events in the breeding lines comingfrom the GSP recurrent selection program (representing58 % of the lines in the collection), which were mainlyderived from double crosses between multiple parents(Additional file 1). The average distance between two ad-jacent marker positions of the TMap considered in this

study (0.27 cM) was higher than the LD decay rate.Since marker genetic positions may not have covered allthe putative recombination points in the collection,there is still the potential to increase genotyping densityof the collection, to improve QTL detection resolution.Previous meta-QTL analysis of Aphanomyces resistanceattempted to reduce QTL CIs but this reduction was po-tentially over-estimated because of the only partial inde-pendence on populations and environments compared[43]. The MLMM model recently proposed by Segura etal. [11], used cofactors that captured background geneticvariance during genome scan to improve the precisionof cofactor detection as found in Composite IntervalMapping.Comparative analyses of the QTL detected by linkage

and association studies are often not carried out oncomparable genetic distance scales in both studies. Con-sequently, several association mapping loci could corres-pond to a single QTL interval identified by linkagemapping. Ingvarsson & Street [10] reported frequentsplitting of linkage mapping intervals into much smallerassociation mapping intervals. Split association mappingintervals involved in a same larger linkage mappinginterval could correspond to several linked genes con-trolling the trait or to a single imprecisely located regiondue to GxE effects on the phenotype. Our findings sug-gest linkage between multiple genes hypothesis whenclosely located LD blocks were detected from differentvariables, as seen for example for the Ae-Ps7.6 region onLGVII (VII.7-8 and VII.13-16 blocks). In other cases theresults were consistent with single genes hypothesiswhen neighbouring blocks were detected for the sametype of variable, as in the Ae-Ps3.1 region on LGIII(III.3–4 for Field_RRI variables; III.6–7 or III.8–9 forField_ADI variables).

GWA study provides new tools for pyramidingAphanomyces resistance alleles in breeding programsThis study identified marker haplotypes at consistentAphanomyces resistance loci, the pyramiding of whichwas associated with increased levels of partial resistancein pea lines of the collection. At the 14 consistent LDblocks identified, 22 favourable haplotypes carrying thefavourable alleles at disease trait-associated markerswere significantly associated with enhanced resistancelevels. The highest resistant lines of the pea-Aphanomyces collection carried a significantly highernumber of favourable haplotypes. The five most resistantlines with no missing haplotypes were breeding linesfrom the AeD99OSW GSP program, all derived fromdirect crosses between the three most partially resistantgermplasm studied in previous QTL analysis [(90–2131x PI180693) x 552]. These five breeding lines had ahigher level of partial resistance than their individual

Desgroux et al. BMC Genomics (2016) 17:124 Page 15 of 21

parents. They combined between eight and 13 of thefavourable haplotypes selected at the 14 consistent re-sistance loci, whereas each of their single parental linescombined between three to nine favourable haplotypes.From this study, different combinations of haplotypesseemed to be related to enhanced levels of resistance.The best combinations were quite difficult to identifysince they could vary depending on the lines. However,the favourable haplotypes defined at several blocks (I.5,II.1, III.3, III.4, III.6 and III.9) appeared to be frequentlyrepresented in the most resistant lines.Our findings suggested that the pyramiding of Apha-

nomyces resistance alleles with moderate to low effectsis a powerful strategy to develop pea lines with increasedlevels of partial resistance to A. euteiches. In the past 30years, such a pyramiding strategy, based on phenotypicrecurrent selection schemes, was successfully developedfor root rot diseases in public USA pea breeding pro-grams [31, 32]. In particular, Lewis and Gritton [92] de-veloped a phenotypic recurrent selection protocolconsisting of one cycle per year, including selection of F2lines on aerial symptoms of the plants in an infestedfield nursery, followed by intercrosses and selfing of thebest lines. The 552 germplasm line was selected fromthe eighth cycle of recurrent selection (Roux-Duparque,Pers. Comm.), suggesting that this line accumulated re-sistance alleles. Accumulation of resistance alleles in the552 germplasm line was demonstrated in this study(nine favourable haplotypes identified). More recently,Lavaud et al. [44] reported that the combination of twoor three of the main consistent Aphanomyces resistanceQTL identified in [43] could increase the partial resist-ance level, depending on genetic backgrounds. In severalother pathosystems, combinations of resistance alleleswere shown to successfully enhance resistance levels, in-cluding stem rust resistance in wheat [93, 94], spot leafdiseases resistance in spring wheat [95] and spot blotchresistance in barley [96].We also identified negative linkages between resistance

and morphological or developmental alleles, to be con-sidered with caution for application in dry pea breeding.Undesirable linkages were previously reported betweenAphanomyces resistance and coloured flowers, long in-ternodes, normal leaves and late-flowering [43], and havebeen avoided in breeding programs [32]. In the presentstudy, associations were highlighted between resistanceand late flowering alleles at markers in the consistentdisease LD blocks VII.13, VII.16 and VII.18. In non-consistent disease LD blocks, undesirable links were alsohighlighted between resistance and normal leaves (LDblock I.4) or late flowering (LD block II.2). In contrast,high resolution of association mapping in this study alsoenabled linkage between resistance and coloured flowers(A gene) alleles to be broken in the genomic region

corresponding to the previous main resistance QTL Ae-Ps2.1 (LD block II.1). Thus, different pea lines of the col-lection carrying favourable haplotypes in the LD blockII.1 have white flowers. However, in the most resistantlines of the collection, such as AeD99OSW- 45-8-7, somenegative associations could not be broken, as those linespresented normal leaves and some of them still hadcoloured flowers. Due to the high resolution of GWAstudy, we could also hypothesize that loci controlling re-sistance and delayed leaf senescence (SGR gene) [83] inthe LD block I.5 are distinct since they were not in LD(r2 > 0.2). Morphological and developmental-trait-lociwhich are not associated with Aphanomyces resistancecould also be detected, suggesting that resistance couldbe improved without including undesirable traits. Forheight and earliness, the height locus III.17 on the Legene and the earliness loci II.2, II.3, II.5,VII.5 and VII.17were identified and not associated with resistance loci.Genetic associations between plant disease resistanceand undesired developmental traits such as late-flowering have been commonly reported in the literature[2]. Association genetic studies could have confirmedsuch negative linkages [97] but sometimes allowed themto be broken, as observed in our study

GWA study identified relevant putative candidate genesunderlying Aphanomyces resistance QTLThe high resolution of the present association mappingstudy and the recent availability of pea transcriptome se-quences connected to the 13,204 SNPs used in thisstudy, constitute unprecedented advantages for identify-ing putative candidate genes underlying Aphanomycesresistance QTL. Indeed, the detection of plant heightQTL in previously cloned genes involved in plantgrowth, and the proximity of flowering genes to earlinessLD blocks suggests that the resolution of the presentGWA study was high enough to pinpoint relevant candi-date genes.GWA study identified a high frequency (64 %) of puta-

tive candidate genes corresponding to a diversity ofstress-related protein functions, underlying resistance-associated intervals. To date, the low number of resist-ance QTL cloned suggested that the molecular functionsunderlying them could be diverse [2, 3, 98]. Hypotheticalgene-encoded proteins underlying resistance QTL in-cluded genes involved in plant-pathogen recognition,signal transduction, activation or repression of transcrip-tion, defence responses and developmental or morpho-logical modifications with pleiotropic effects. Such adiverse range of genes underlying resistance intervalswas also identified in this study. Putative candidategenes underlying major effect QTL Ae-Ps4.4-4.5, re-cently validated for pea resistance to the American A.euteiches strain Ae109 [44] (LD block IV.8), included

Desgroux et al. BMC Genomics (2016) 17:124 Page 16 of 21

basic helix-loop-helix (bHLH) and ethylene-responsivetranscription factors and a NBS-LRR. A second main ef-fect QTL Ae-Ps7.6, validated for pea resistance to theFrench A. euteiches strain RB84 (LD blocks VII.8 andVII.10), included two putative infection-induced pro-teins. Some different types of putative stress-relatedgenes clustered in the disease LD blocks identified inthis study and thus may be strong candidates for under-lying one of the known mechanisms. For example, LDblocks III.9 and I.5 contained six and four putative can-didate genes, respectively, involved in mechanisms ofpathogen recognition, transcription regulation, signaltransduction and/or defence response.Other LD blocks seemed to be interesting as they are

very close to NBS-LRR cloned genes (III.14, IV.10,V.I andVI.1), however they were detected in single environmentsand had moderate effects. Genes encoding NBS-LRR pro-tein domains are known to be clustered in crop genomes,such as described in Medicago truncatula [99], since theyare known to be subjected to rapid evolution throughlocal duplications. In pea, Tayeh et al. [49] recently re-ported a large cluster of 14 NBS-resistance-like genes atthe bottom of LGIII. This genomic region, including SNPmarkers assigned to these 14 NBS-like genes, co-localizedwith disease LD block III.14 and is close to consistent LDblock III.15 identified in this study.Some disease LD blocks contained putative genes

similar to those underlying disease resistance QTLcloned in plants, such as ABC transporters (III.6) andNBS-LRR proteins (I.5, IV.8 and VII.3)). An ABC trans-porter protein have been shown to underlie the Lr34major QTL controlling durable resistance to leaf ruststripe rust and powdery mildew in wheat [100]. OtherLD blocks could also be detected close to cloned symbi-osis genes (III.2, IV.8, VI.5 and VII.11), suggesting pos-sible common mechanisms between plant-pathogensand plant-beneficial organisms interactions [101]. SomeLD blocks also included plant or root developmentgenes (IV.3, IV.7, VII.7 and VII.10), suggesting their in-volvement in disease resistance, tolerance or avoidanceas reported for several pathosystems [100].

ConclusionIn the present GWA study, most of the previously iden-tified Aphanomyces resistance QTL were validated witha finer resolution than before, by taking advantage of thenew plant and genomic resources developed in pea. Newpoints of discussion were raised regarding comparison ofassociation and linkage analysis to dissect polygenic dis-ease resistances in plants. This study provides new toolsfor breeding, including pea germplasm and SNP markersassociated with resistance, as well as useful informationabout marker haplotypes at main resistance loci and un-desirable allele linkages with resistance. It demonstrates

that pyramiding resistance alleles is a key strategy for in-creasing levels of partial resistance to Aphanomyces rootrot in pea. The choice of resistance loci to be pyramidedremains to be explored, especially using NILs [44]. Thisstudy also identified relevant candidate genes to be con-firmed and validated in future studies. These will benefitfrom further genomic resources in progress, developedin part from a complete pea reference genome sequencebecoming available, for increasing resolution of theGWA study.

Availability of supporting dataFurther data sets supporting the results of this articleare included in Additional files 16 and 17.

Additional files

Additional file 1: Description of the pea-Aphanomyces collection. (a)Line categories, as referred in Fig. 1b and Additional file 9 in legends; (b) Linecodes in the Biological resource center (CRB) database of INRA Dijon, France;(c) Country codes ISO3166-1; (d) Cv: Cultivar, Bl: breeding line, Gmp: germ-plasm, Lv: local variety, Wa: wild accession; (e) Fd: fodder, Gd: garden; (f) Ws:winter sown, Ss: spring sown; (g) Nl: normal leaves, af: afila leaves, (h) W: white,Pu: purple, Pi: pink; (i) Wr: wrinkled, Sm: smooth; #: parents of GSP breedingprograms; $: parents of RIL populations. (XLSX 29 kb)

Additional file 2: Simple sequence repeat (SSR) markers used togenotype the pea-Aphanomyces collection. (a) main Aphanomycesresistance quantitative trait loci (QTL) defined in [43]. (XLSX 14 kb)

Additional file 3: Example of MLMM R package plot outputs foranalysis of the MFA.Dim.1 variable. (A): Bonferroni correction of thehighest marker p-value for each step of the forward and backwardanalysis. Optimal step (dashed blue vertical line) is determined as thelargest stepwise mixed model regression in which all cofactors have –log10(p-value) above the mBonf threshold (dashed black horizontal line).(B): QQ-plot at the optimal step. (C): Partition of variance for each forwardand backward step. Variance explained by population structure (PCA;grey); all markers used as cofactors (blue); Kinship matrix (green) andunexplained variance (= missing heritability; red). The dashed blue verticalline represents the optimal step according to mBonf threshold. (D)Genome-wide Manhattan plot without marker cofactors. The dotted linerepresents the multiple-Bonferroni threshold above which markers areconsidered as significant. (E): Genome-wide Manhattan plot at theoptimal step with marker cofactors (4 red dots). The dotted linerepresents the multiple-Bonferroni threshold above which markers areconsidered as significant. (PDF 148 kb)

Additional file 4: Statistical analysis results of Aphanomycesresistance, earliness and height scoring data in the pea-Aphanomyces collection. (a) CC: controlled condition experiment, Field:Infested field experiment, RRI: Root rot index, ADI1: First aerial declineindex, ADI2: Second aerial decline index, FLO1: date to 50 % bloom,FLO2: date to 100 % bloom, RIPE: date to 100 % dried plants, HT: heightof plants, 09 to 13: year of experiment, All: global variables over all fieldenvironments, DI: Dijon-Epoisses, Burgundy, France, RI: Riec-sur-Belon,Brittany, France, REN: Rennes-Le Rheu, Brittany, France, KEN: Kendrick, ID,USA, ‘$’ Residual normality statistics are not exact but graphs of residualsare correct, (b) Mean, minimum, maximum and standard error in thepea-Aphanomyces collection; (c) ANOVA significance p-value codes:0 < ‘***’ < 0.001≤ ‘**’ < 0.01≤‘*’ < 0.05≤ ‘ns’ < 1; (d): Genotype effect; (e):Replicate effect; (f): Environment effect; (g): Genotype x Environmentinteraction effect; (h) mean-based heritability. (XLSX 16 kb)

Additional file 5: Frequency distribution of least square meansobtained for Aphanomyces resistance, earliness and plant height inthe pea-Aphanomyces collection. Least square means were obtainedfrom analysis of variance for three Aphanomyces resistance traits (Root

Desgroux et al. BMC Genomics (2016) 17:124 Page 17 of 21

Rot Index, first and second Aerial Decline Indexes, coded RRI, ADI1 andADI2, respectively), three earliness traits (dates to 50 % bloom, 100 %bloom and 100 % dried plants, coded FLO1, FLO2 and RIPE, respectively)and plant height (HT). Aphanomyces resistance traits were assessed overnine infested field environments and against two reference strains of A.euteiches (RB84 and Ae109) in controlled conditions. Earliness and heightwere assessed in two healthy environments. n: total number of pea linesassessed; m: mean ± standard deviation of the pea-Aphanomycescollection; h2: mean-based heritability. (PDF 38 kb)

Additional file 6: Clustered heatmap of correlation coefficientsbetween disease resistance, earliness and height variables data.Variables are coded as described in Additional file 4. Only correlationswith a p-value < 0.05 are shown. Colours represent level of correlation,coded as follow: dark red: r < −0.8; medium red: −0.8 < r < −0.5; lightred: −0.5 < r < −0.2; white: −0.2 < r < 0.2 or p-value > 0.05; light green:0.2 < r < 0.5; medium green: 0.5 < r < 0.8; dark green: r > 0.8. Clusteringis based on the UPGMA method. (PDF 48 kb)

Additional file 7: Variable contribution in the two first MultipleFactor Analysis (MFA) principal components. (a) Variables are codedas described in Additional file 4, (b) percentage of variable contributionin the construction of the MFA axis, ‘*’: the six highest contributions; (c)cos2 indicates the representation quality of the variable on the MFA axis,‘*’: 0.2 < cos2 < 0.5, ‘**’: cos2 > 0.5; (d) correlation between variable andMFA axis, ‘*’: 0.5 < r2 < 0.8, ‘**’: r2 > 0.8. (XLSX 17 kb)

Additional file 8: Linkage disequilibrium (LD) decay in thepea-Aphanomyces collection. Coloured curves represent the estimatedLD decay for each linkage group (LG). Dashed horizontal lines representhalf of the maximum LD value (LD threshold=0.5). Arrows represent theLD decay rate, as the estimated genetic distance (cM) to drop to the LDthreshold on each LG. (PDF 157 kb)

Additional file 9: Population structure of the pea-Aphanomyces col-lection based on Principal Component Analysis (PCA). PCA fromGAPIT R package [75] based on 2937 markers. Distributions of pea lines ofthe collection are represented on the first three principal components(A–C), which explain a total of 20.91 % of inertia (D). Categories of pealines are described in Fig. 1 and Additional file 1. (D) Inertia contributionof each principal component (from PC1 to PC169). (PDF 151 kb)

Additional file 10: Clustered heatmap of the Kinship matrix. Kinshipmatrix from the GAPIT R package [75] based on 2937 SNP markers.Clustering is based on the UPGMA method. Colours represent the degreeof relationship between two given lines. Pea line information is describedas in Additional file 1. Lines sharing same pedigree or end use or sowingtype are well clustered thus the Kinship matrix efficiently represents therelationships between individuals. (PDF 1253 kb)

Additional file 11: Genetic loci associated with Aphanomycesresistance, earliness and plant height detected by genome-wide as-sociation mapping using the MLMM method. (a) Genetic positions areestimated on the genetic map resulting from the projection of themarker map from Hamon et al. [43] onto the consensus map from Tayehet al. [49]; (b) A linkage disequilibrium (LD) block is designed with allmarkers in LD (r2 > 0.2) with the associated marker. LD blocks are namedin consecutive numerical order following their linkage group (LG) nameand their genetic position; (c) Aphanomyces resistance and earlinessMeta-QTL (quantitative trait loci) described in [43]; (d) Initial QTL forAphanomyces resistance and earliness before meta-analysis described in[41, 43]; (e) Trait-associated marker(s) and their genetic position(s) in theLD blocks. PsCam- markers are SNPs developed by Tayeh et al. [49], Af isa morphological gene involved in leaves shape, other markers are SSRs(detected allele in brackets); (f) Minor allele frequency (MAF) of eachdetected marker in the collection; (g) Variables are coded as described inAdditional file 4; (h) significant p-value of the marker-trait association <2.5E-5; (i) mean value of the trait corrected with the effect of all markersincluded in the MLMM analysis; (j) allelic effect of the marker on the vari-able, the sign depends on the allele code; (k) standard error at the markeraround the mean value of the trait in the collection. (XLSX 32 kb)

Additional file 12: Information on the connectivity betweenprojected maps. (a) HMap: consensus map from Hamon et al. [43]; (b)TMap: consensus map from Tayeh et al. [49]; (c) number of common

markers between HMap and TMap for each linkage group (LG); (d):Number of inversions of marker positions on each LG; (e) number ofinversions of marker positions in confidence intervals of the quantitativetrait loci (QTL) described by Hamon et al. [43]. (XLSX 11 kb)

Additional file 13: Description of linkage disequilibrium (LD) blockhaplotypes in the pea-Aphanomyces collection. (a) LD blocks are namedas described in Additional file 11; (b) Each LD block is subdivided into markerhaplotypes according to its allelic composition. Marker haplotypes are namedwith the LD block name followed by an Arabic numeral. Only haplotypeswithout missing values or heterozygous markers are shown; For each LDblock: (c) list of markers significantly detected by genome-wide association inthe LD block; (d) the first line is a list of the markers included in the LD block(detected markers in bold font and their markers in (LD) in plain font); (e) thesecond line is the genetic positions of the listed markers on the genetic mapdescribed in Additional file 11; (f) following lines are the pairwise LD r2 valuesbetween each marker defined in the LD block and the marker detected byGWA; (g) For SNP markers, ‘AA’ and ‘BB’ are allele codes of Genome Studioanalysis while for SSR markers, amplicon lengths are indicated. For each Apha-nomyces resistance associated marker, enhancing allele(s) is in bold dark greenfont and unfavourable allele(s) in bold dark red font; (h) the first line is the listof the variables significantly associated with a marker in the LD block, variablesare described as in Additional file 4; (i) next lines are p-values of the marker-trait associations; (j) grey or coloured lines are mean phenotypic values ±standard error for marker haplotypes carried by more than 5 % of the linesfrom the pea-Aphanomyces collection, letters indicate significantly differentmeans (Tukey-HSD, α < 0.05); dark green and red haplotypes: favourable andunfavourable haplotypes of the LD block, respectively, as defined in theMateriel and Methods section of the manuscript. (XLSX 73 kb)

Additional file 14: Marker haplotype composition of the pea-Aphanomyces collection. (a) Name of marker haplotype carried by eachline as described in Additional file 13, at each of the 14 consistent linkagedisequilibrium (LD) blocks detected; dark green and red: favourable andunfavourable marker haplotypes, respectively, as defined in the Materialand Methods section of the manuscript; (b) coordinates of each line ofthe collection on the first Multiple Factor Analysis principal component;the pea lines of the collection are ranked according to these coordinates;(c) Composition of each line in favourable and unfavourable haplotypes,is described in Additional file 13; (d) marker haplotypes with either amissing value or heterozygous score for at least one marker of a LDblock. (XLSX 66 kb)

Additional file 15: Annotation of markers in detected linkagedisequilibrium (LD) blocks for putative candidate genes. (a) LD blockwere constructed with markers detected by genome-wide association map-ping and markers in LD with them, LD block are named according to Add-itional file 11; (b) genetic positions are estimated as described in Additionalfile 11; For each LD block:(c) traits associated with markers in the LD block;(d) Pairwise LD between each marker and the marker(s) detected by GWAin the LD block (detected marker in bold); (e) putative protein annotationunderlying each marker based on comparative Blast X data from Tayeh etal. [49], see Methods section; (f) putative function of the protein in plant dis-ease resistance based on the literature; (g) putative function of the proteinin plant development based on the literature; (h) reference for protein func-tion involved in plant disease resistance or plant development; (i) Peacloned genes referred in literature; (j) extract of annotation data from Tayehet al. [49], Blast X of marker sequence onto P. sativum, M. truncatula, G. maxand A. thaliana proteins. (XLSX 131 kb)

Additional file 16: Least square means (LSMeans) of phenotypicdata in the pea-Aphanomyces collection after ANOVA. Variables arecoded as in Additional file 4. NA: missing data. (XLSX 71 kb)

Additional file 17: Genotyping data of the pea-Aphanomyces col-lection used for the GWA analysis before filters and imputation. NA:missing data. (a) Morphological gene genotyping data are coded “AA”when phenotypes are white flowers, normal leaves or smooth seeds forA, Af and R genes, respectively, and “BB” when they are colored flowers,afila leaves or wrinkled seeds; (b) PCR specific genotyping data are coded“AA” or “BB” when specific to the first or second allele; (c) SSR genotypingdata are presented with the allele size (in bp); (d) SNP genotyping dataare coded “AA” or “BB” when homozygous for the first or second allele,“AB” when heterozygous. (XLSX 7252 kb)

Desgroux et al. BMC Genomics (2016) 17:124 Page 18 of 21

Abbreviations(c)DNA: (coding) deoxyribonucleic acid; ADI: aerial decline index; ADI1: firstaerial decline index; ADI2: second aerial decline index; CC: controlledconditions; CI: confidence interval; DI: Dijon-Epoisses, France; DSP: Dark SkinPerfection (germplasm); E: environment; EMMA: efficient mixed-model associ-ation; FLO1: number of calendar days to 50 % bloom; FLO2: number ofcalendar days to 100 % bloom; G: genotype; GAPIT: Genome Association andPrediction Integrated Tool; GSP: Groupement des sélectionneurs de poisprotéagineux; GWA: genome-wide association; GxE: Genotype x environmentinteraction; HMap: consensus map from [43]; HT: plant height at 100 %bloom; INRA: Institut national de la recherche agronomique; KEN: Kendrick,USA; LD: linkage disequilibrium; LG: linkage group; LM: linear model;LRR: leucine rich repeat; LSMeans: least square means; MAF: minor allelefrequency; MAGIC: multi-parent advanced generation inter-cross;MAS: marker assisted selection; mBonf: multiple-Bonferroni threshold;MFA: multiple factors analysis; MFA.Dim.1: first principal component of MFA;MFA.Dim.2: second principal component of MFA; MLMM: multi-locus mixedmodel; NAM: nested association mapping; NBS: nucleotide binding site;NGS: next generation sequencing; NIL: near-isogenic line; PC: principalcomponent; PCA: principal components analysis; PCR: polymerase chainreaction; QTL: quantitative trait locus; R: replicate; REN: Rennes-Le Rheu,France; RI: Riec-sur-Belon, France; RIL: recombinant inbred line; RIPE: numberof calendar days to 100 % dried plants; RRI: root rot index; SGR: stay greengene; SNP: single nucleotide polymorphism; SSR: simple sequence repeats;THMap: consensus map from projection of HMap onto TMap;TMap: consensus map from [49]; UPGMA: unweighted pair group methodwith arithmetic mean; USA: United States of America; USDA: United StatesDepartment of Agriculture.

Competing interestsThe authors declare that they have no competing interests.

Authors’ contributionsAD participated in phenotypic data acquisition, performed all the phenotypicand GWA analyses and drafted the manuscript. VLA contributed togenotyping data acquisition (SSR), association analysis and to the draftmanuscript. MRD managed the GSP breeding program and constructedbreeding lines included in the pea-Aphanomyces collection. GA coordinatedthe SNP genotyping, performed its analysis and contributed to geneticanalysis. NT provided the genetic map, gene annotation and contributed togenetic analysis. AM contributed to phenotypic data acquisition and to thedesign of the pea-Aphanomyces collection. JPR and PV coordinated andparticipated in field experiments at Riec-sur-Belon and Rennes-Le Rheu. PMcoordinated and participated in field experiments at Dijon-Epoisses. CJC andRJMG designed field Aphanomyces assays in the USA and participated infield experiments at Kendrick. CP prepared the DNA for SNP genotyping. JBdesigned and coordinated the Infinium® BeadChip genotyping of thepea-Aphanomyces collection. AB participated to the design of the GSPbreeding scheme and of the pea-Aphanomyces collection. MMD and VBparticipated in the conception of the study and in drafting the manuscript.MLPN coordinated the overall study and the manuscript draft. All authorsread and approved the final manuscript.

AcknowledgementsThis work was supported by the pre-doctoral fellowship of Aurore Desgroux, fromMAP (Ministère de l’Agriculture, de l’Agroalimentaire et de la Forêt, Paris, France). Itwas funded by Terres Univia (Paris, France) for phenotyping and SSR genotypingdata and for the engineer fellowship of Virginie L’Anthoëne, and by INRA (Méta-programme SELGEN, France) (CropDL project) and the ANR project (ANR-09-GENM-026) for SNP genotyping data. We greatly acknowledge GSP for havingprovided breeding lines improved for Aphanomyces resistance. We warmly thankthe INRA experimental units of Dijon-Epoisses and Rennes-Le Rheu, as well as UNI-LET (Union Nationale Interprofessionnelle des Légumes Transformés), Quimperlé,France, for their contribution to the field experiments. We greatly acknowledgeSolène Coëdel from the Genouest genotyping platform and Marie-Christine LePaslier from the Evry genotyping platform, for their contribution to the genotyp-ing of the pea-Aphanomyces collection. We are very grateful to Mathieu Siol andAnne-Sophie Bouchet, who provided help in genome-wide association analysisand to Gwenola LeRoy who helped in the controlled conditions experiments. Wegreatly thank Clément Lavaud for his critical review of the manuscript. We thankLeigh Gebbie for revising the English written style of the manuscript.

Author details1INRA, UMR IGEPP 1349, Institut de Génétique et Protection des Plantes,Domaine de la Motte au Vicomte, BP 35327, 35653 Le Rheu Cedex, France.2INRA, UMR 1347 Agroécologie, 17 rue de Sully, 21065 Dijon Cedex, France.3GSP, Domaine Brunehaut, 80200 Estrées-Mons Cedex, France. 4PISOM, UMTINRA/Terres Inovia, UMR IGEPP 1349, Domaine de la Motte au Vicomte, BP35327, 35653 Le Rheu Cedex, France. 5Terres Inovia, 11 rue de Monceau, CS60003, 75378 Paris Cedex, France. 6INRA, Domaine Expérimental d’Epoisses,UE0115, 21110 Bretenières Cedex, France. 7USDA, ARS, Grain LegumeGenetics and Physiology Research Unit, Pullman, WA 99164-6434, USA.8USDA, ARS, Western Regional Plant Introduction Station, Washington StateUniversity, Pullman, WA 99164-6402, USA. 9AgroCampus Ouest, UMR IGEPP1349 IGEPP, 65 rue de Saint Brieuc, 35042 Rennes Cedex, France. 10PresentAddress: Nestlé R&D Center Tours, 101 Avenue Gustave Eiffel, 37097 ToursCedex 2, France. 11Present Address: Chambre d’Agriculture de l’Aisne, 1 rueRené Blondelle, 02007 Laon Cedex, France.

Received: 31 July 2015 Accepted: 2 February 2016

References1. St Clair DA. Quantitative disease resistance and quantitative resistance Loci

in breeding. Annu Rev Phytopathol. 2010;48:247–68. doi:10.1146/annurev-phyto-080508-081904.

2. Poland JA, Balint-Kurti PJ, Wisser RJ, Pratt RC, Nelson RJ. Shades of gray: theworld of quantitative disease resistance. Trends Plant Sci. 2009;14(1):21–9.doi:10.1016/j.tplants.2008.10.006.

3. Kou Y, Wang S. Broad-spectrum and durability: understanding ofquantitative disease resistance. Curr Opin Plant Biol. 2010;13(2):181–5.doi:10.1016/j.pbi.2009.12.010.

4. Boyd LA, Ridout C, O’Sullivan DM, Leach JE, Leung H. Plant-pathogeninteractions: disease resistance in modern agriculture. Trends Genet. 2013;29(4):233–40. doi:10.1016/j.tig.2012.10.011.

5. Huang X, Han B. Natural variations and genome-wide association studies incrop plants. Annu Rev Plant Biol. 2014;65:531–51. doi:10.1146/annurev-arplant-050213-035715.

6. Korte A, Farlow A. The advantages and limitations of trait analysis withGWAS: a review. Plant Methods. 2013;9:29. doi:10.1186/1746-4811-9-29.

7. Gupta PK, Kulwal PL, Jaiswal V. Association mapping in crop plants:opportunities and challenges. Adv Genet. 2014;85:109–47. doi:10.1016/B978-0-12-800271-1.00002-0.

8. Bao Y, Vuong T, Meinhardt C, Tiffin P, Denny R, Chen S et al. Potential ofassociation mapping and genomic selection to explore PI 88788 derivedsoybean cyst nematode resistance. Plant Genome. 2014;7(3).doi:10.3835/plantgenome2013.11.0039.

9. Cheng P, Holdsworth W, Ma Y, Coyne CJ, Mazourek M, Grusak MA et al.Association mapping of agronomic and quality traits in USDA pea single-plant collection. Mol Breed. 2015;35(2). doi:10.1007/s11032-015-0277-6.

10. Ingvarsson PK, Street NR. Association genetics of complex traits in plants.New Phytol. 2011;189(4):909–22. doi:10.1111/j.1469-8137.2010.03593.x.

11. Segura V, Vilhjalmsson BJ, Platt A, Korte A, Seren U, Long Q, et al. Anefficient multi-locus mixed-model approach for genome-wide associationstudies in structured populations. Nat Genet. 2012;44(7):825–30.doi:10.1038/ng.2314.

12. Sauvage C, Segura V, Bauchet G, Stevens R, Do PT, Nikoloski Z, et al.Genome-wide association in tomato reveals 44 candidate loci for fruitmetabolic traits. Plant Physiol. 2014;165(3):1120–32. doi:10.1104/pp.114.241521.

13. Begum H, Spindel JE, Lalusin A, Borromeo T, Gregorio G, Hernandez J, et al.Genome-wide association mapping for yield and other agronomic traits inan elite breeding population of tropical rice (Oryza sativa). PLoS One. 2015;10(3), e0119873. doi:10.1371/journal.pone.0119873.

14. Zhao X, Han Y, Li Y, Liu D, Sun M, Zhao Y, et al. Loci and candidate geneidentification for resistance to Sclerotinia sclerotiorum in soybean (Glycinemax L. Merr.) via association and linkage maps. Plant J. 2015;82(2):245–55.doi:10.1111/tpj.12810.

15. Samayoa LF, Malvar RA, Olukolu BA, Holland JB, Butron A. Genome-wideassociation study reveals a set of genes associated with resistance to theMediterranean corn borer (Sesamia nonagrioides L.) in a maize diversitypanel. BMC Plant Biol. 2015;15:35. doi:10.1186/s12870-014-0403-3.

Desgroux et al. BMC Genomics (2016) 17:124 Page 19 of 21

16. Myles S, Peiffer J, Brown PJ, Ersoz ES, Zhang Z, Costich DE, et al. Associationmapping: critical considerations shift from genotyping to experimentaldesign. Plant Cell. 2009;21(8):2194–202. doi:10.1105/tpc.109.068437.

17. Pascual L, Albert E, Sauvage C, Duangjit J, Bouchet J-P, Bitton F, et al.Dissecting quantitative trait variation in the resequencing era:complementarity of bi-parental, multi-parental and association panels. PlantSci. 2015. doi:10.1016/j.plantsci.2015.06.017.

18. Yu J, Holland JB, McMullen MD, Buckler ES. Genetic design and statisticalpower of nested association mapping in maize. Genetics. 2008;178(1):539–51.

19. Benson JM, Poland JA, Benson BM, Stromberg EL, Nelson RJ. Resistance togray leaf spot of maize: genetic architecture and mechanisms elucidatedthrough nested association mapping and near-isogenic line analysis. PLoSGenet. 2015;11(3), e1005045. doi:10.1371/journal.pgen.1005045.

20. Cavanagh C, Morell M, Mackay I, Powell W. From mutations to MAGIC:resources for gene discovery, validation and delivery in crop plants. Curr OpinPlant Biol. 2008;11(2):215–21. http://dx.doi.org/10.1016/j.pbi.2008.01.002.

21. Pascual L, Desplat N, Huang BE, Desgroux A, Bruguier L, Bouchet JP, et al.Potential of a tomato MAGIC population to decipher the genetic control ofquantitative traits and detect causal variants in the resequencing era. PlantBiotechnol J. 2015;13(4):565–77. doi:10.1111/pbi.12282.

22. Zila CT, Ogut F, Romay MC, Gardner CA, Buckler ES, Holland JB. Genome-wide association study of Fusarium ear rot disease in the U.S.A. maizeinbred line collection. BMC Plant Biol. 2014;14:372. doi:10.1186/s12870-014-0372-6.

23. Hao D, Cheng Y, Chen G, Lu H, Shi M, Zhang Z, et al. Identification ofsignificant single nucleotide polymorphisms for resistance to maize roughdwarf disease in elite maize (Zea mays L.) inbred lines. Euphytica. 2014;203(1):109–20. doi:10.1007/s10681-014-1277-z.

24. FAOSTAT. FAOSTAT data base. Food and agriculture organisation of theUnited Nations - statistic division. 2015. http://faostat3.fao.org/home/E.Accessed 17 janv 2016.

25. McPhee K. Dry pea production and bredding - A mini-review. J Food AgricEnviron. 2003;1(1):64–9.

26. Jones FR, Drechsler C. Root rot of peas caused by Aphanomyces euteiches.J Agric Res. 1925;30(4):293–325.

27. Papavizas GC, Ayers WA. Aphanomyces species and their root diseases inpea and sugarbeet: a review. Washington, USA: United States Dept ofAgriculture; 1974. Technical bulletin n°1485:1–158.

28. Wicker E, Rouxel F. Specific behaviour of French Aphanomyces euteichesDrechs. populations for virulence and aggressiveness on pea, related toisolates from Europe, America and New Zealand. Eur J Plant Pathol.2001;107:919–29.

29. Wicker E, Moussart A, Duparque M, Rouxel F. Further contributions to thedevelopment of a differential set of pea cultivars (Pisum sativum) toinvestigate the virulence of isolates of Aphanomyces euteiches. Eur J PlantPathol. 2003;109:47–60.

30. Chatterton S, Bowness R, Harding MW. First report of root rot of field peacaused by Aphanomyces euteiches in Alberta, Canada. Plant Dis.2015;99(2):288. doi:10.1094/PDIS09140905PDN.

31. Gritton ET. Registration of five pea root rot resistant germplasm lines ofprocessing pea. Crop Sci. 1990;30:1166–7.

32. Kraft JM. Registration of 90–2079, 90–2131, and 90–2322 pea germplasms.Crop Sci. 1992;32:1076.

33. Davis DW. MN 144, MN 313, and MN 314: garden pea lines resistant to rootrot caused by Aphanomyces euteiches Drechs. HortSci. 1995;30(3):639–40.

34. Malvick DK, Percich JA. Identification of Pisum sativum germ plasm withresistance to root rot caused by multiple strains of Aphanomyces euteiches.Plant Dis. 1999;83(1):51–4.

35. Kraft JM, Coffman VA. Registration of 96–2052, 96–2058, 96–2068, 96–2198,and 96–2222 pea germplasms. Crop Sci. 2000;40:301–2.

36. Pilet-Nayel L, Coyne CJ, Hamon C, Lesne A, Le Goff I, Esnault R et al., editors.Understanding genetics of partial resistance to Aphanomyces root rot inpea for new breeding prospects. 3rd International Aphanomyces Workshopon Legumes; 2007 07–09 November; Rennes, Brittany, France.

37. Roux-Duparque M, Boitel C, editors. Breeding peas for resistance toAphanomyces root rot: current main outputs of three breeding programes.5th Eur Conf Grain Legumes; 2004 7-11th June; Dijon, France.

38. Moussart A, Devaux C, Muel F, Pilet-Nayel M, Baranger A, Tivoli B et al.,editors. Improving partial resistance to Aphanomyces root rot in GSPbreeding program. 3rd International Aphanomyces Workshop on Legumes;2007 07–09 November; Rennes, Brittany, France.

39. Pilet-Nayel ML, Muehlbauer FJ, McGee RJ, Kraft JM, Baranger A, Coyne CJ.Quantitative trait loci for partial resistance to Aphanomyces root rot in pea.Theor Appl Genet. 2002;106(1):28–39. doi:10.1007/s00122-002-0985-2.

40. Pilet-Nayel ML, Muehlbauer FJ, McGee RJ, Kraft JM, Baranger A, Coyne CJ.Consistent quantitative trait loci in pea for partial resistance toAphanomyces euteiches isolates from the United States and France.Phytopathology. 2005;95(11):1287–93. doi:10.1094/PHYTO-95-1287.

41. Hamon C, Baranger A, Coyne CJ, McGee RJ, Le Goff I, L’Anthoene V, et al.New consistent QTL in pea associated with partial resistance toAphanomyces euteiches in multiple French and American environments.Theor Appl Genet. 2011;123(2):261–81. doi:10.1007/s00122-011-1582-z.

42. McGee RJ, Coyne CJ, Pilet-Nayel ML, Moussart A, Tivoli B, Baranger A, et al.Registration of pea germplasm lines partially resistant to aphanomyces rootrot for breeding fresh or freezer pea and dry pea types. J Plant Regist. 2012;6(2):203–7. doi:10.3198/jpr2011.03.0139crg.

43. Hamon C, Coyne CJ, McGee RJ, Lesné A, Esnault R, Mangin P, et al. QTLmeta-analysis provides a comprehensive ciew of the moderately lowdiversity of loci controlling partial resistance to Aphanomyces euteiches infour pea sources of resistance. BMC Plant Biol. 2013;13:45.

44. Lavaud C, Lesné A, Piriou C, Le Roy G, Boutet G, Moussart A, et al. Validationof QTL for resistance to Aphanomyces euteiches in different pea geneticbackgrounds using near isogenic lines. Theor Appl Genet. 2015.doi:10.1007/s00122-015-2583-0.

45. Duarte J, Riviere N, Baranger A, Aubert G, Burstin J, Cornet L, et al.Transcriptome sequencing for high throughput SNP development and geneticmapping in pea. BMC Genomics. 2014;15:126. doi:10.1186/1471-2164-15-126.

46. Sindhu A, Ramsay L, Sanderson L-A, Stonehouse R, Li R, Condie J, et al.Gene-based SNP discovery and genetic mapping in pea. Theor Appl Genet.2014;127(10):2225–41. doi:10.1007/s00122-014-2375-y.

47. Leonforte A, Sudheesh S, Cogan NO, Salisbury PA, Nicolas ME, Materne M,et al. SNP marker discovery, linkage map construction and identification ofQTLs for enhanced salinity tolerance in field pea (Pisum sativum L.). BMCPlant Biol. 2013;13:161. doi:10.1186/1471-2229-13-161.

48. Boutet G, Alves-Carvalho S, Falque M, Peterlongo P, Lhuillier E, Bouchez Oet al. SNP discovery and genetic mapping using Genotyping by Sequencingof whole genome genomic DNA from a pea RIL population. BMCGenomics. 2016;17:121. doi:10.1186/s12864-016-2447-2.

49. Tayeh N, Aluome C, Falque M, Jacquin F, Klein A, Chauveau A, et al.Development of two major resources for pea genomics: the GenoPea 13.2KSNP array and a high density, high resolution consensus genetic map. PlantJ. 2015. doi:10.1111/tpj.13070.

50. Kwon S-J, Brown AF, Hu J, McGee R, Watt C, Kisha T, et al. Genetic diversity,population structure and genome-wide marker-trait association analysisemphasizing seed nutrients of the USDA pea (Pisum sativum L.) core collection.Genes Genomics. 2012;34(3):305–20. doi:10.1007/s13258-011-0213-z.

51. Diapari M, Sindhu A, Warkentin TD, Bett K, Tar’an B. Population structureand marker-trait association studies of iron, zinc and seleniumconcentrations in seed of field pea (Pisum sativum L.). Mol Breed. 2015;35(1).doi:10.1007/s11032-015-0252-2.

52. Burstin J, Salloignon P, Chabert-Martinello M, Magnin-Robert JB, Siol M,Jacquin F, et al. Genetic diversity and trait genomic prediction in a peadiversity panel. BMC Genomics. 2015;16:105. doi:10.1186/s12864-015-1266-1.

54. Moussart A, Wicker E, Duparque M, Rouxel F, editors. Development of anefficient screening test for pea resistance to Aphanomyces euteiches. 4thEuropean conference on grain legumes; 2001 8–12 July; Cracow, Poland.

55. Aubert G, Morin J, Jacquin F, Loridon K, Quillet MC, Petit A, et al. Functionalmapping in pea, as an aid to the candidate gene selection and forinvestigating synteny with the model legume Medicago truncatula. TheorAppl Genet. 2006;112(6):1024–41. doi:10.1007/s00122-005-0205-y.

56. Loridon K, McPhee K, Morin J, Dubreuil P, Pilet-Nayel ML, Aubert G, et al.Microsatellite marker polymorphism and mapping in pea (Pisum sativum L.).Theor Appl Genet. 2005;111:1022–31. doi:10.1007/s00122-0050014-3.

57. Doyle JJ, Doyle JL. Isolation of plant DNA from fresh tissue. Focus.1990;12:13–5.

58. Alves-Carvalho S, Aubert G, Carrère S, Cruaud C, Brochot A-L, Jacquin F,et al. Full-length de novo assembly of RNA-seq data in pea (Pisum sativum L.)provides a gene expression atlas and gives insights into root nodulation in thisspecies. Plant J. 2015;84:1–19. doi:10.1111/tpj.12967.

59. R Core Team. R: A language and environment for statistical computing. RFoundation for Statistical Computing, Vienna, Austria. 2014. http://www.R-project.org/. Accessed 13 Jul 2015.

Desgroux et al. BMC Genomics (2016) 17:124 Page 20 of 21

60. de Mendiburu F. agricolae: Statistical Procedures for Agricultural Research. Rpackage version 1.2-1. 2014. http://CRAN.R-project.org/package=agricolae.Accessed 13 Jul 2015.

61. Hervé M. RVAideMemoire: Diverse basic statistical and graphical functions. Rpackage version 0.9-36. 2014. http://CRAN.R-project.org/package=RVAideMemoire. Accessed 13 Jul 2015.

62. Lenth RV, Hervé M. lsmeans: Least-Squares Means. R package version 2.15.2015. http://CRAN.R-project.org/package=lsmeans. Accessed 13 Jul 2015.

63. Revelle W. psych: Procedures for Psychological, Psychometric, andPersonality Research. R package version 1.5.1. 2015. http://CRAN.R-project.org/package=psych. Accessed 13 Jul 2015.

64. Benjamini Y, Hochberg Y. Controlling the false discovery rate - A practicaland powerful approach to multiple testing. J R Stat Soc series B StatMethodol. 1995;57(1):289–300.

65. Warnes GR, Bolker B, Bonebakker L, Gentleman R, Liaw WHA, Lumley T et al.gplots: Various R Programming Tools for Plotting Data. R package version 2.16.0. 2015. http://CRAN.R-project.org/package=gplots. Accessed 13 Jul 2015.

66. Husson F, Josse J, Le S, Mazet J. FactoMineR: Multivariate Exploratory DataAnalysis and Data Mining. R package version 1.29. 2015. http://CRAN.R-project.org/package=FactoMineR. Accessed 13 Jul 2015.

67. Husson F, Josse J. missMDA: Handling missing values with/in multivariatedata analysis (principal component methods). R package version 1.7.3. 2014.http://CRAN.R-project.org/package=missMDA. Accessed 13 Jul 2015.

68. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al.PLINK: a tool set for whole-genome association and population-basedlinkage analyses. Am J Hum Genet. 2007;81:559–75.

69. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-generation PLINK: rising to the challenge of larger and richer datasets.Gigascience. 2015;4:7. doi:10.1186/s13742-015-0047-8.

70. Purcell SM, Chang CC. PLINK 1.9. 2015. https://www.cog-genomics.org/plink2. Accessed 13 Jul 2015.

71. Schwender H, Fritsch A. scrime: Analysis of High-Dimensional CategoricalData such as SNP Data. R package version 1.3.3. 2013. http://CRAN.R-project.org/package=scrime. Accessed 13 Jul 2015.

72. Schwender H. Imputing missing genotypes with weighted k nearestneighbors. J Toxicol Environ Health. 2012;75(8–10):438–46. doi:10.1080/15287394.2012.674910.

73. Sved JA. Linkage disequilibrium and homozygosity of chromosomesegments in finite populations. Theor Popul Biol. 1971;2(2):125–41. http://dx.doi.org/10.1016/0040-5809(71)90011-6.

74. Huang X, Wei X, Sang T, Zhao Q, Feng Q, Zhao Y, et al. Genome-wideassociation studies of 14 agronomic traits in rice landraces. Nat Genet. 2010;42(11):961–7.

75. Lipka AE, Tian F, Wang Q, Peiffer J, Li M, Bradbury PJ, et al. GAPIT: genomeassociation and prediction integrated tool. Bioinformatics. 2012;28(18):2397–9. doi:10.1093/bioinformatics/bts444.

76. Buckler Lab for Maize Genetics and Diversity. GAPIT source code andtutorial. 2012. http://www.maizegenetics.net/#!gapit/cmkv. Accessed 27november 2015.

77. Segura V. Multi-Locus Mixed-Model (MLMM) script. 2013. https://sites.google.com/site/vincentosegura/mlmm. Accessed 13 Jul 2015.

78. Hothorn T, Bretz F, Westfall P. Simultaneous inference in general parametricmodels. Biom J. 2008;50(3):346–63. doi:10.1002/bimj.200810425.

79. Sosnowski O, Charcosset A, Joets J. BioMercator V3: an upgrade of geneticmap compilation and quantitative trait loci meta-analysis algorithms.Bioinformatics. 2012;28(15):2082–3.

80. Voorrips RE. MapChart: software for the graphical presentation of linkagemaps and QTLs. J Hered. 2002;93(1):77–8.

81. Timmerman-Vaughan GM, Frew TJ, Weeden NF. Characterization and linkagemapping of R-gene analogous DNA sequences in pea (Pisum sativum L.). TheorAppl Genet. 2000;101(1–2):241–7. doi:10.1007/s001220051475.

82. Prioul-Gervais S, Deniot G, Receveur EM, Frankewitz A, Fourmann M,Rameau C, et al. Candidate genes for quantitative resistance toMycosphaerella pinodes in pea (Pisum sativum L.). Theor Appl Genet. 2007;114(6):971–84. doi:10.1007/s00122-006-0492-y.

83. Armstead I, Donnison I, Aubry S, Harper J, Hörtensteiner S, James C, et al.Cross-species identification of Mendel’s I locus. Science. 2007;315(5808):73.

84. Afzal AJ, Wood AJ, Lightfoot DA. Plant receptor-like serine threonine kinases:roles in signaling and plant defense. Mol Plant Microbe Interact. 2008;21(5):507–17. doi:10.1094/mpmi-21-5-0507.

85. Walter S, Kahla A, Arunachalam C, Perochon A, Khan MR, Scofield SR, et al. Awheat ABC transporter contributes to both grain formation and mycotoxintolerance. J Exp Bot. 2015;66(9):2583–93. doi:10.1093/jxb/erv048.

86. Hecht V, Foucher F, Ferrandiz C, Macknight R, Navarro C, Morin J, et al.Conservation of Arabidopsis flowering genes in model legumes. PlantPhysiol. 2005;137(4):1420–34. doi:10.1104/pp.104.057018.

87. Weller JL, Liew LC, Hecht VFG, Rajandran V, Laurie RE, Ridge S, et al. Aconserved molecular basis for photoperiod adaptation in two temperatelegumes. Proc Natl Acad Sci U S A. 2012;109(51):21158–63.

88. Hofer J, Turner L, Hellens R, Ambrose M, Matthews P, Michael A, et al.UNIFOLIATA regulates leaf and flower morphogenesis in pea. Curr Biol.1997;7(8):581–7. doi:10.1016/s0960-9822(06)00257-0.

89. Weston DE, Elliott RC, Lester DR, Rameau C, Reid JB, Murfet IC, et al. The PeaDELLA proteins LA and CRY are important regulators of Gibberellin synthesis androot growth. Plant Physiol. 2008;147(1):199–205. doi:10.1104/pp.108.115808.

90. Reinecke DM, Wickramarathna AD, Ozga JA, Kurepin LV, Jin AL, Good AG,et al. Gibberellin 3-oxidase gene expression patterns influence Gibberellinbiosynthesis, growth, and development in pea. Plant Physiol. 2013;163(2):929–45. doi:10.1104/pp.113.225987.

91. Warkentin TD. Pea. In: De Ron AM, editor. Grain legumes. Handbook ofplant breeding, vol. 10. New York: Springer; 2015.

92. Lewis ME, Gritton ET. Use of one cycle of recurrent selection per year forincreasing resistance to Aphanomyces root rot in peas. J Am Soc Hortic Sci.1992;117(4):638–42.

93. Letta T, Maccaferri M, Badebo A, Ammar K, Ricci A, Crossa J, et al. Searchingfor novel sources of field resistance to Ug99 and Ethiopian stem rust racesin durum wheat via association mapping. Theor Appl Genet. 2013;126(5):1237–56. doi:10.1007/s00122-013-2050-8.

94. Naruoka Y, Garland-Campbell KA, Carter AH. Genome-wide associationmapping for stripe rust (Puccinia striiformis F. sp. tritici) in US PacificNorthwest winter wheat (Triticum aestivum L.). Theor Appl Genet. 2015;128(6):1083–101. doi:10.1007/s00122-015-2492-2.

95. Gurung S, Mamidi S, Bonman JM, Xiong M, Brown-Guedira G, Adhikari TB.Genome-wide association study reveals novel quantitative trait Lociassociated with resistance to multiple leaf spot diseases of spring wheat.PLoS One. 2014;9(9), e108179. doi:10.1371/journal.pone.0108179.

96. Zhou H, Steffenson B. Genome-wide association mapping reveals geneticarchitecture of durable spot blotch resistance in US barley breedinggermplasm. Mol Breed. 2013;32(1):139–54. doi:10.1007/s11032-013-9858-4.

97. Goudemand E, Laurent V, Duchalais L, Tabib Ghaffary S, Kema GJ, Lonnet P,et al. Association mapping and meta-analysis: two complementaryapproaches for the detection of reliable Septoria tritici blotch quantitativeresistance in bread wheat (Triticum aestivum L.). Mol Breed. 2013;32(3):563–84. doi:10.1007/s11032-013-9890-4.

98. Michelmore RW, Christopoulou M, Caldwell KS. Impacts of resistance genegenetics, function, and evolution on a durable future. Annu RevPhytopathol. 2013;51(1):291–319. doi:10.1146/annurev-phyto-082712-102334.

99. Ameline-Torregrosa C, Wang B-B, O’Bleness MS, Deshpande S, Zhu H, Roe B,et al. Identification and characterization of nucleotide-binding site-leucine-rich repeat genes in the model plant Medicago truncatula. Plant Physiol.2008;146(1):5–21. doi:10.1104/pp.107.104588.

100. Costes E, Lauri PE, Simon S, Andrieu B. Plant architecture, its diversity andmanipulation in agronomic conditions, in relation with pest and pathogenattacks. Eur J Plant Pathol. 2013;135(3):455–70. doi:10.1007/s10658-012-0158-3.

101. Podolich O, Ardanov P, Zaets I, Pirttilä A, Kozyrovska N. Reviving of theendophytic bacterial community as a putative mechanism of plantresistance. Plant Soil. 2015;388(1–2):367–77. doi:10.1007/s11104-014-2235-1.

Desgroux et al. BMC Genomics (2016) 17:124 Page 21 of 21